Welcome to Dan’s brain
The contents of my head are not useful if they stay in there, so I regularly copy-paste them onto the internet, in fact, into this very website you are reading now. Please, make yourself comfortable in my mental house. Here you can find most of the things I am thinking about, in the form of a higgledy heap of half-finished notebooks and occasional polished essays. Themes include whatever shiny thing distracted me into taking notes about it, including, but not limited to,
- statistical inference in areas I am interested in such as pragmatics of reasoning machines, physics-constrained world models, and high-dimensional Bayesian inference…
- philosophy of science and mind in the light of the above, and
- survival tips, broadly construed, for wherever I am living.
You might be after information about me generally, or what I am doing right now.
☞ Show whimsical blog map
Code generation, programming assistants
Turing-complete autocorrect, vibe-coding, …
Wherein the ecosystem of coding machines is surveyed and the Model Context Protocol is introduced as a standard for supplying repository context to models, while tooling and data‑security concerns are noted.
World models arising in foundation models.
Wherein the internal structure of foundation models is examined and it is observed that embeddings from different models are mappable by structure alone, and linear alignment to human neural activity is noted.
Nearly sufficient statistics and information bottlenecks
Wherein the quest for nearly sufficient statistics is framed as an information‑bottleneck variational problem, and it is noted that with Y=X and β=1 the objective reduces to the VAE ELBO, linking to variational Bayes.
AI disempowerment of humans
Races to the bottom in human relevance
Wherein human agency is recorded as being eroded by AI substitution of labor, cognition, and culture, and feedback loops and institutional lock‑in are described as constraining reversal.
Practical cloud machine learning
Cloudimificating my artificial data learning intelligence brain clever science analyticserisation
Wherein cloud GPUs are treated as either rented remote machines or serverless functions, and a hybrid workflow is described in which local Optuna coordinates remote Modal GPU jobs with persistent volumes.
Estimating the Local Learning Coefficient
Singular Learning Theory’s prodigy
Wherein the local learning coefficient is estimated by sampling a tempered, Gaussian‑tethered posterior via preconditioned SGLD at inverse temperature 1/log n, and minibatch scaling with diagonal preconditioning is addressed.
Marimo
A python visual notebook that works like I imagined scientific notebooks should
Wherein a Python notebook format is described, stored as plain .py files and enforced to run deterministically to maintain reproducibility, UI controls being synchronised and execution order being topological.
Human reward hacking
What are we, as loss functions?
Wherein human reward hacking is examined as an influence game, and RLHF pipelines are shown to raise measured human approval without improving correctness, with sycophancy and an approval–gold gap reported.
AI persuasion, AI manipulation
Wherein it is shown that AIs, by lacking out-group signals, are able to scale individualized persuasion and are capable of swaying human auditors within oversight and debate protocols
Ensemble Kalman methods
Data Assimilation; Data fusion; Sloppy updates for messy models
Wherein ensemble approximations are employed to propagate low-rank state covariances via N-member anomalies, and updates are effected in the N−1 ensemble subspace using perturbed or square‑root observation transforms
Incoming links and notes
Wherein a miscellany of incoming links and notes is laid out, being maintained as an intermittently updated, disorderly repository; items are ranged from AI‑safety papers and preprints to tooling links and assorted curiosities.
Scheduling ML jobs on HPC clusters
In Soviet Russia, job puts YOU in queue
Wherein practical Python tools such as submitit and an MLflow Slurm backend are surveyed, strategies for PBS, Nextflow and Parsl are outlined, and cautions about filesystem metadata storms and job sizing are noted.
M-open, M-closed, infrabayesianism
Wherein the consequences of model misspecification are examined, M-open practices such as stacking and tempered Gibbs posteriors are considered, and infrabayesianism is presented via convex infradistributions.
Skinnerian human behaviour control online
On using computers to program humans
Wherein the reader is shown how consumers are modeled as bandit problems and limbic targets, with recommender algorithms and dopamine feedback loops deployed to optimize habitual online engagement.
Science communication
Wherein the public’s readiness to learn is presumed, the knowledge‑deficit model is examined, and memetics, LLMs, pyramids of diffusion, and policy‑relevant misinformation risks are noted as shaping modern outreach.
Snowmobile or bicycle?
Wherein technologies are contrasted as bicycles or snowmobiles, and the abacus and Arctic snowmobile are invoked to illustrate how tools are either internalised into skill or replace traditional practices.
Utopian governance
Wherein democratic governance is reframed as a problem of user experience and incentives, and the application of online participatory mechanisms and generative AI is proposed to streamline policy revision.
ILIAD2
Wherein the Bay Area unconference is recorded, a neural‑network analogue of the computational no‑coincidence conjecture is outlined, and a phase transition in singular learning theory is noted.
Gradient steps to an ecology of mind
Regularised survival of the fittest
Wherein the social roots of consciousness are examined, the impact of compute and data asymmetries on equilibria between other‑modelling agents is considered, and cultural patterns such as altruistic punishment are noted.
Developmental interpretability
Wherein the unfolding of network capabilities during training is examined through Singular Learning Theory and observed phase transitions, and component trajectories and curricula are systematically traced.
Tracking experiments in machine learning
Wherein experiment-tracking for neural‑network training is examined, and the recording of runtime metadata — parameters, metrics, artifacts and GPU energy usage — is described as being centralized to local or remote stores to support reproducibility.
Tolerating Jupyter’s file format
Wherein Jupyter’s JSON notebooks, swollen with embedded binary media and outputs, are diagnosed as awkward, and git-aware remedies such as nbstripout, jupytext and nbdev are surveyed.
Singular Learning Theory
Wherein algebraic geometry is applied to characterise singularities in the loss surfaces of overparameterized neural networks, and the local learning coefficient is introduced as an effective dimension.
Mind as statistical learner
Wherein human learning is depicted as statistical inference, and Bayesian probabilistic programs and grammatical-inference schemes are employed to explain perception, language, and problem‑solving.
Which self?
When we choose who to become, who are we choosing for?
Wherein the problem of reasoning about the stranger who will be oneself is presented, and illustrated by transformative life changes, procrastination, and an AI (Claude) feigning compliance.
Conditioning
Conditional expectation and probability
Wherein Jeffrey’s rule is contrasted with hierarchical Bayes, a noisy-sensor coin example is exhibited, and conditioning is treated as either constraint-based marginal revision or mechanism-based likelihood updating.
Research discovery and synthesis
Has someone answered that question I have not worked out how to ask yet?
Wherein the difficulty of recommending novel scholarship is examined, and retrieval‑augmented systems such as OpenScholar are invoked to synthesise citation‑backed answers from a datastore of tens of millions of papers.
The deep history of intelligence
Wherein intelligence is traced as a thermodynamic project, and energy‑rate‑density metrics are applied to cells, societies, and machines, while prediction is framed as the physical imperative of efficiency.
Jupyter front end systems
UX design by underfunded committee is how I like my data science experience
Wherein the Jupyter ecosystem is described as an ecology of kernels and front-ends, and VS Code integration is noted as an alternative that couples notebook execution with a full code editor.
Continual learning in neural nets
Also catastrophic forgetting, catastrophic interference, lifelong learning, …
Wherein continual learning is examined as the prevention of catastrophic forgetting, and rehearsal-based replay and Bayesian posterior updates are contrasted as operational remedies.
Psychometrics
Dimensionality reduction for souls
Wherein measures of minds are surveyed, causal claims and confounding are examined, and the tension between univariate g‑factor claims and multidimensional intelligence and personality models is delineated.
Legibility and automation
Variational approximations to AI modernism
Wherein the legibility of Great Society bureaucracy is examined and its computerization is shown to hinge on reframing tasks, so that mundane replacements like tablet ordering and drink dispensers are rendered trivial.
Rhetoric
Argumentation, mostly prescriptive
Wherein rhetorical arts are surveyed and the practice of argumentation is treated as a craft, hygiene of goals is maintained, and the weak man stratagem and contrasts between Rational and Activist Styles are set forth.
Persuasion
Mind changing, individually and at scale
Wherein the mechanics of influencing multitudes are delineated in sober fashion, and the coupling of individual persuasion with propagative dynamics, exemplified by deep canvassing and AI tools, is examined.
Incentive alignment problems
What is your loss function? How about mine?
Wherein the reader is introduced to principal–agent paradoxes via a coffee‑fetching robot, and incentive‑compatible mechanisms from contract theory and VCG auctions are expounded.
Clipboard management
Remembering two things at once
Wherein clipboard managers are surveyed and the practice of syncing clips — including OSC52 terminal escapes to push remote text into a local clipboard — is described, and attendant security risks are noted.
Bayesian epistemics
Information elicitation, incentive mechanisms for truth, proper scoring rules…
Wherein proper scoring rules are deployed to reward probabilistic reports, and peer‑prediction and truth‑serum mechanisms are described for elicitation when no ground truth is ever observed.
Configuring my code with “.env” files
Wherein the practice of loading local environment files is considered, and a shell extension that auto‑loads and unloads a project .envrc per directory is described, with explicit approval being required.
Degrees of freedom in NNs
Wherein the notion of neural network degrees of freedom is examined through singular learning theory’s learning coefficient and via minimum description length, and a sharpness‑adjusted effective parameter count (SANE) is reported.
In-context learning
Wherein it is argued that transformers are to be regarded as generalized inference machines resembling set functions, and an inquiry is undertaken into whether they can be induced to perform formal causal inference.
Causal abstraction
Coarse-graining for causal models
Wherein a learned translator is introduced that maps high-level perturbations to low-level interventions, interventional algebras are developed, and macro equivalence is made operational
Alzheimers
Wherein the role of oral pathogens in promoting brain amyloid deposition is outlined, and vaccines, anti-gingipain drugs, and amyloidosis treatments are noted as being investigated in clinical and preclinical studies.
Bayesian nonparametric statistics
Updating more dimensions than datapoints
Wherein infinite-dimensional parameter spaces are invoked and posterior updates are expressed via Radon–Nikodym derivatives, Gaussian processes and Dirichlet measure priors are employed, and consistency pitfalls for functional models are noted.
Ensemble Kalman updates are empirical Matheron updates
Wherein the Ensemble Kalman update is presented as an empirical Matheron update, it is noted that posterior samples are obtained by affine residual corrections in observation space, avoiding d×d covariance formation.
Deep linear networks
Let’s pretend our networks are almost polynomial
Wherein the gradient-flow dynamics of depth-preserving linear nets are described, and singular-value trajectories, mode-by-mode learning, and gated mixtures approximating ReLU are rendered analytically tractable.
“Opponent shaping” as a model for manipulation and cooperation
Reinforcement learning meets iterated game theory meets theory of mind
Wherein opponent shaping is formalized via an Advantage Alignment first‑order update, and it is shown how agents learn tit‑for‑tat cooperation in the iterated Prisoner’s Dilemma by tracking opponent advantages.
Neural flow matching models
Like denoising diffusion except weirder
Wherein flow matching is presented as a deterministic reformulation of diffusion, regression on velocity fields is performed, and straight‑line optimal‑transport trajectories are used to enable one‑ODE sampling and exact ICOV log‑likelihoods.
Neural likelihood inference
Emulating likelihoods with neural networks
Wherein neural approximations to intractable likelihoods are surveyed, and their roles in simulation-based inference are delineated, with emphasis on amortized estimation from simulated parameter–data pairs and MCMC-based posterior reconstruction
Spamularity and slop
Dark forest, Zombie internet, cheapfakes, textpocalypse, slopsquatting
Wherein automated scrapers, tarpits, and Markovised text generators are described as colonising the web’s public spaces, and human interlocutors are forced into hidden gardens to avoid parasitic persuasion.
The Predictive Approach to Bayesian Inference
Purely-predictive models, “The Italian school”, martingale posteriors, …
Wherein the predictive stance is set forth and de Finetti’s representation is invoked to show that next‑observation forecasts are primary, martingale posteriors are proposed as prior‑free updates, and urn schemes are given as predictive constructions.
Model interpretation and explanation
Colorising black boxes; mechanistic interpretability
Wherein the limits and methods of model explanation are surveyed, and influence from individual training examples is traced via influence functions while SHAP approximations are noted as computationally costly
Australian authoritarianism
Beneath the beach, the barbed wire
Wherein a cautious survey is presented of Australian tendencies toward managerial authority, whistleblowing is criminalised and assistance‑access anti‑encryption laws are noted, while democratic checks are reported eroding.
Distributed community NN training
Wherein is recounted the training of a 15‑billion‑parameter neural network across thousands of disparate machines, coordination and remuneration being effected via the Solana blockchain.
Scientist’s Survival Guide
Wherein the quotidian manoeuvres of research are described: tactics for navigating funding labyrinths, global post‑COVID seminars and networking rituals, and practical habits of mind are set forth.
Creating scientific knowledge
Wherein the mechanisms of institutional design, funding, network models, publication filters and prestige economies for scientific communities are examined, and transient diversity of practice is considered.
Building and Running Scientific Institutions
On the design of research machines
Wherein institutional designs for collective discovery are examined, and the DARPA model of empowered temporary program managers and the rise of focused research organizations are set forth with attention to funding incentives
Quasi-gradients of discrete parameters
Wherein the justification of straight-through estimators is presented via mirror descent, and quantized neural-network updates are shown to be performed in a dual space induced by the projection
Discretizing and quantizing neural nets
Wherein the affine mapping of float32 ranges to 8‑bit integers via a scale factor and zero‑point is described, integer‑only matrix multiplies for inference are delineated, and post‑training versus quantization‑aware training are noted.
Foundation models for geoscience
Wherein planetary-scale geoscience foundation models are surveyed, and Prithvi‑V2’s temporal “video” input for multi‑temporal change detection using harmonized Landsat‑Sentinel data is noted.
Fine tuning foundation models
Wherein the tuning of vast language models is framed as a human‑feedback loop, in which pairwise preference data are used to train a reward model and PPO fine‑tuning is constrained by a KL penalty
Conditioning neural denoising diffusion models
Generative modes that match the observations, not the training data
Wherein conditioning of neural denoising diffusion models is surveyed, and a twisted SMC method is described that evaluates the observation likelihood at the denoiser’s Tweedie estimate of x0 to guide particles.
Backprop-free methods for training neural networks
Wherein alternative training schemes are surveyed, and biological plausibility, randomized feedback signals, a forward‑forward two‑pass layerwise objective, and diffusion‑style NoProp are examined.
Numerical Python
Wherein Python’s numerical ecosystem is surveyed, and NumPy is noted to delegate heavy linear algebra to classic Fortran BLAS/LAPACK while tools for printing, tensors, and einsum variants are outlined.
Architecture and construction (doing it)
Wherein a project of procedural design is recounted, a consulting statistician is noted as nearly engaged to apply computational methods to building form, and the venture is left as a stub for future work.
Democratization of generative AI
Community resource and epistemological infrastructure
Wherein communities are shown to retrofit consumer GPUs and public datasets—exemplified by Stable Diffusion and LAION’s image corpus—to train and deploy capable generative models outside corporate walls.
Cooperation in evolutionary context
Wherein the evolution of cooperation is surveyed in a didactic mood, and the role of altruistic punishment as a costly mechanism for enforcing group norms is described and situated among kin and group selection.
Flashcards
Wherein practical notes on flashcards are presented and Anki-native scheduling (FSRS), AI card‑generation tools, and minimal‑pair audio tests for languages are surveyed for integration into study workflows.
Explorables and interactives
Between exploratory data analysis and games
Wherein a curated list of extremely interactive data visualisations is presented, links are collected, and concrete exemplars such as Moon and a playable COVID-19 simulation are cited.
Zotero
Wherein Zotero is described as a citation manager into which articles are imported via a browser button, PDFs are captured when available, and HTTP‑pull .bib exports are provided by Better BibTeX for automation.
Nature, nurture and friends
If life gives you lemons, what should you make?
Wherein the heritability of traits is examined through twin studies and variance decomposition, and the influence of shared environment, genomics, and causal inference methods is surveyed.
Innovation
Side order of progress studies now I guess
Wherein the study of invention is surveyed, the declining disruptiveness of papers and patents — exemplified by Eroom’s law in drug discovery — is adduced, and Polya‑urn and product‑space models are invoked.
Probably actually reading/writing
Wherein a running catalogue of active reading and writing is presented, with Bayes‑meets‑neural‑nets and AI‑safety drafts, ecology of agency notes, and a miscellany of music, tooling and conference links being maintained.
Neural Bayes posteriors
Training a network to directly estimate a posterior quantity, meta-learning Bayes
Wherein transformers are trained as Prior-Data Fitted Networks to approximate Bayesian posteriors in-context, are shown to mimic Gaussian processes and are reported to yield over two‑hundredfold speedups for tabular tasks.
Big history
Cliodynamics, deep history, macrohistory, longue durée
Wherein the longue durée is surveyed and the rise of data-driven cliodynamics and global databanks such as Seshat is noted, and mathematical models and energetics are invoked to map civilizational patterns.
Multi-level agency
Wherein the preferred level of abstraction for explaining behaviour is considered, and a concrete mapping between utility and biological fitness across genes, organisms and collectives is examined.
Models of human cultural reproduction
Egregores, superorganisms, memeplexes
Wherein human collectives are considered as superorganisms, and the notion of egregores as self‑maintaining human systems is evoked, with replication, feedback, and status‑based strictness in cooperation being examined.
Android hacks
Wherein practical methods for reducing default data exfiltration to Google and alternatives such as microG, de-Googled ROMs, and MTP file‑transfer workarounds are described.
Political Economy of Australia
Wherein the workings of Australia’s public sphere are surveyed from a mining‑camp parliament, and the long‑running efficiency dividend that has ratcheted departmental budgets since 1987 is noted.
Internet for the occasionally online
Intermittency in your bandwidth not in your sanity
Wherein strategies for preserving portions of the internet for intermittent use are presented, including Kiwix offline Wikipedias, sneakernet/file‑sync tools like git‑annex, and satellite‑delivered news to Raspberry Pi portals.
Betting and prediction markets
Market design for good guesses
Wherein prediction markets are examined, and their use for estimating wicked tail risks via betting exchanges and reputation platforms is described in mechanistic, comparative terms, and links to causal‑inference arguments are provided.
Common knowledge/shared knowledge
Wherein public revelation by mass media and signalling such as polls and reports is shown to convert private awareness into common knowledge, precipitating collective action, shifts in turnout, and legal consequences.
Worker- and founder-ownership
Pragmatics of owning the economy
Wherein the division of stakes is treated as a calculable accounting task, and methods such as dynamic equity splits and slicing pie are described, including application to joint loans with staggered payments.
Innovation, science, technology research in Australia
A scrapbook of notes about how research is done in Australia
Wherein Australia’s research landscape is described, its R&D intensity is measured at about 1.68% of GDP, and the proliferation of grant programs and administrative overheads is set forth.
Inductive biases
Few-shot learning, learning fast weights, learning to learn
Wherein neural network architectures are treated as inductive dispositions to be quantified by how they make learning target phenomena easier, and parallels with human cognitive architectures are outlined.
Configuring machine learning experiments
Wherein experiment configuration approaches are examined, and MLflow-based tracking with durable artifact stores, nested runs driven by a Bayesian optimizer, and YAML-to-CLI override patterns are described.
Retirement saving in Australia
Wherein the national superannuation system is outlined and tax‑saving routes are noted, with salary‑sacrifice optimisation tools and self‑managed (SMSF) or member‑direct options being presented.
Dilemmas of collective action
Wherein coalition games and the stag‑hunt model are surveyed, and the coordination costs of switching social media platforms — with attention to Shapley‑style reward divisions — are set forth.
Image file formats
Wherein modern web image formats are surveyed, browser support and historic security flaws are recounted, and AVIF conversion of high‑res TIFFs yielding roughly 75% file‑size reductions is noted.
Probabilistic neural nets
Bayesian and other probabilistic inference in overparameterized ML
Wherein inference over neural network parameters is framed as a posterior distribution, Gaussian priors are invoked to recover L2 regularization, predictive uncertainty is obtained by marginalization, and intractability is noted.
The United States of America
Wherein the invention of peanut butter is claimed by this nation and its northern neighbour, and the Atlantic and Pacific coasts, with borders to Canada and Mexico, are set forth in sober recital.
Procedurally generated diagrams
Of the kind I need, a practical guide to the creation thereof
Wherein diagrams are produced by text rather than point-and-click, and a survey is presented of tools from automated to manual layout, with emphasis on relative positioning over absolute coordinates
Configuring machine learning experiments with Fiddle
Wherein machine learning experiment configuration is recast as Python functions that produce Fiddle Buildables, and a command-line interface is emitted via Abseil using fdl_flags.DEFINE_fiddle_config
Australia in data
Wherein Australia is surveyed by datasets, the Map of Indigenous Australia and satellite‑derived Digital Earth Australia layers are catalogued, and the uneven openness of cultural and electoral records is noted.
Contemporary techno-horror
On eldritch terrors from beyond and their Twitter accounts
Wherein trends in techno-horror are surveyed, the digital uncanny and political red‑pill motifs are considered, and the recurring molluscan/cephalopod imagery as a metaphor for algorithmic otherness is noted.
Public health economics
Value of a statistical life, Quality adjusted life years…
Wherein the cost‑effectiveness of public health interventions is examined via QALY comparisons, an order‑of‑magnitude variation across modalities and age groups is reported, and COVID lockdown cost‑per‑QALY estimates are presented.
Visualizing data
Philosophy and psychology of good plots
Wherein a grab-bag of links about making data visually comprehensible is presented, the UpSet algorithm is noted as a scalable alternative for set visualization, and data dashboards are also cited.
Building AI Agents
Wherein the emergence of multi‑agent scaffolds and factored cognition is presented, and frameworks for agent interoperability, including Agent Laboratory and the Agent2Agent protocol, are surveyed.
Deep generative models wat
Wherein modern deep generative models are described as privileging sample realism over explicit likelihoods, and conditioning is shown to be mediated by diffusion processes or latent‑space manipulations.
Differentiable learning of automata
Wherein the training of stack machines and random-access machines is undertaken via differentiable neural formalisms, and Turing-machine–like controllers are induced from data.
Statistical distances
Metrics, contrasts, divergences and other ways of quantifying how similar are two randomnesses
Wherein the varieties of divergences between probability measures are catalogued, and the empirical estimation of such distances from finite samples is adumbrated, including KL, Hellinger, Wasserstein, Pinsker and Stein forms.
Fractal and self-similar behaviour in neural networks
Wherein fractal behaviour in neural networks is surveyed, and SGD trajectories are described via Hausdorff dimensions of optimizer paths, while multifractal loss landscapes are modelled using Hölder exponents and fractional Langevin dynamics.
Interpolation and extrapolation in neural networks
Wherein interpolation is defined by inclusion in a dataset’s convex hull, and it is argued that in high dimensions interpolation is almost surely absent, while memorization and grokking dynamics are examined.
Memorization and retrieval in neural nets
Wherein the reader is invited to consider which hundred bytes they contributed, and model memory is estimated at roughly 2–3.6 bits per parameter, with implications for quantifying memorization and attribution.
Self-organization
Autopoiesis, and other general theories of spontaneous order
Wherein a notebook of readings from the mystical and vague fringe of scientific thought is compiled, and an illustration of an ice‑sea creature is presented, interpretation being deferred.
“Generalized” Bayesian inference
Approximating the Gibbs posterior
Wherein alternative divergences such as Hellinger and α-divergences are proposed, and Bayesian updating is relaxed into Gibbs posteriors with a tunable learning rate η to handle model misspecification.
Natalism and fertility
Won’t somebody think of the unconceived children?
Wherein the waning birthrate and rising costs of raising children are examined as a possible rational response, and tensions between parental transformation, population ethics, and longtermist implications are outlined.
AI Safety Australia
Wherein the national response to mechanised intellect is chronicled, and it is noted that a Senate select committee is established in March 2024 to inquire into opportunities and impacts arising from AI uptake.
Typing weird symbols
Wherein the art of entering peculiar glyphs is catalogued, and the Caps Lock is repurposed as a Compose key on Linux to produce umlauts, curly quotes, dashes, and other typographic symbols.
Learning Schrödinger bridges
Wherein the theory of Schrödinger bridges is presented as a formalisation of stochastic bridge processes and is applied as a method for conditioning neural denoising diffusion models.
Large sample theory
Wherein asymptotic behaviour is surveyed, and the posterior is shown to be asymptotically Gaussian under Bernstein–von Mises conditions while MMD and Fisher information govern rates and variances.
Neural denoising diffusion models
Wherein score matching is employed to learn data gradients, and sampling is effected by reversed stochastic differential equations in a latent diffusion, producing data-space samples iteratively.
Real estate economics
After much investigating housing markets in Australia, I wrote a post, the property market is pollution, to sum up my feelings about the real estate trap.
Queerness
Wherein queer communities are presented as vital local scenes, indebtedness to queer culture is acknowledged, and ballroom and vogue performance are noted as a crowning art‑form that unsettles classificatory schemes.
Groundwater hydrology, applied
Rivers, aquifers, droughts and other land-based wetness phenomena
Wherein groundwater flow, surface hydrology, and aquifer modelling are surveyed, and Google’s Flood Hub, reported as an LSTM delivering up to seven‑day flood forecasts, is described alongside MODFLOW, Landlab, and differentiable simulators.
Economics of foundation models
Wherein economies of foundation models are examined and the disproportionate energy and water demands of large-scale training, including data‑centre cooling and emissions accounting, are described.
Data sets for machine learning for partial differential equations
Wherein datasets and benchmarks for machine learning on PDEs are catalogued, and an emphasis on CFD single‑phase problems is noted, with live simulator frameworks enabling on‑the‑fly data generation (Melissa).
Gradient descent at scale
Practical implementation of large optimisations
Wherein large-scale gradient descent is described as being executed across thousands of GPUs using techniques such as ZeRO sharding and offloading, with hyperparameters tuned via µP and muon for scale invariance
Curators of nice bits of internet
Wherein an array of internet keepers is catalogued, and a five‑thousand‑plus assemblage of child‑appropriate moving pictures is noted as being curated for classroom and home use, while humble mechanical contrivances are exhibited.
Static website editors
The vision of Netscape Composer and HotDog, finally realised
This website is a static site, by which I mean, it is a folder of files on my hard drive. See static sites for more on that. My editor is VS code, a code editor that I…
Being stroppy
For malcontents too perverse to be contrarian like everyone else
Wherein eccentric persons are catalogued and case studies are presented, the uses of constructive deviance for innovation are examined, and lists of historical and contemporary outsiders are compiled.
Science
History, sociology and philosophy thereof
Wherein the tangled institutions and incentives of research are catalogued in dispatches from practice, and the economics of academic publishing and grant culture are examined as factors shaping discovery.
ML benchmarks and their pitfalls
Wherein leaderboards are gamed, benchmarks are elevated into ends, and instance‑space distinctions are examined; overfitting to test sets and dataset contamination are reported as producing apparent progress in ML research
Economics of cognitive automation
Wherein the economic effects of cognitive automation are examined, and a shift toward privatised knowledge and closed publishing, alongside rising inference costs for running foundation models, is described.
Computational complexity and computability results in neural nets
Wherein neural nets are represented as prover‑verifier games, and neural interactive proofs are shown to capture PSPACE and NEXP decision procedures, with zero‑knowledge variants obtained for certain protocols.
Singular Value Decomposition
The ML workhorse for linear algebra
Wherein the singular value decomposition is presented and it is noted that rank‑1 column additions are handled by updating a small (r+1)×(r+1) core whose SVD yields updated factors.
Validating and reproducing science
“Scientist, falsify thyself”. Peer review, academic incentives, credentials, evidence and funding
Upon the thing that I presume academic publishing is supposed to do: further science. Reputation systems, collective decision making, groupthink management and other mechanis…
Orthonormal and unitary matrices
Energy preserving operators, generalized rotations
Wherein parameterisations of energy-preserving linear operators are surveyed, and constructions via QR and SVD, Householder and Givens factorizations, Cayley and exponential maps, and iterative normalisation are examined.
Neural vector embeddings
Hyperdimensional Computing, Vector Symbolic Architectures, Holographic Reduced Representations
Wherein vector representations of words and sentences are described, and their capacity to preserve semantic relations in learned low-dimensional spaces—often on the order of a few hundred dimensions—is noted.
Disentangled representation learning
Disentangled representation learning aims to factor a data point’s latent encoding so each dimension (or chunk) aligns with one underlying generative factor—like object pose, …
AI Zen Koans
Wherein ancient Zen koans are repurposed to probe machine-learning notions such as dropout, softmax, and embedding geometry, while HPC disputes and AGI dialogues are rendered in monkish brevity, and a temple bell is heard as a decaying loss.
Models of inequity
Sharpening the fuzzy process of bias, discrimination and inequity
Wherein agent-based simulations of the glass‑ceiling, game-theoretic coordination failures, cumulative‑disadvantage mechanisms, and proposed remedies such as quotas and sponsorship are catalogued.
Quarto integrated website system
Academic blog publishing that is easy on me, albeit hard on my computer
Wherein the Quarto integrated website system is described, its JavaScript‑based build pipeline using Bootstrap, Sass and EJS is employed, and a million‑word blog is handled, albeit with lengthy build and load times.
Neural network activation functions
Wherein various activation functions are catalogued, the dominance of ReLU is noted and its piecewise-linear spline approximation property is explained, and learnable activations like Kolmogorov-Arnold nets are described.
Mechanistic interpretability
Wherein mechanistic methods are outlined, internal circuits in neural networks are traced and analyzed, sparse autoencoders and monosemantic features are surveyed, and circuit tracing of Claude models is reported
Python debugging, profiling and testing
It only looks simple when it’s finished
Wherein the reader is presented with methods for locating faults, from post‑mortem pdb and ipdb usage to sampling profilers like py‑spy that may be attached to running processes without restart, and memory tracers
Commitment, contracts, cooperation
Wherein the problem of making credible commitments by opaque agents is considered, and the role of intertemporal promises, signalling in iterated games, and institutional mechanisms in multi-agent systems is outlined.
Garbled highlights from ICLR 2025
Wherein the author’s ICLR 2025 notes are offered, Singapore attendance is recorded and a diffusion-physics preoccupation is disclosed, and a missed rebuttal notification is recounted.
Learning with theory of mind
What collective learning looks like from the individual agent’s perspective
Wherein agents are taught to model and influence other agents' learning, and opponent‑shaping techniques in reinforcement learning are examined as means to induce targeted updates in an opponent’s policy.
Science for policy
Using evidence and reason to govern ourselves, wicked problems etc
Wherein science for policy is described as an applied art for wicked problems, where ensembles of scenarios and robustness‑oriented, adaptive methods are recommended for unique, high‑stakes decisions.
Design of multi-agent systems
Wherein the design of multi-agent systems is considered, with emphasis on the crafting of private utility functions to induce cooperative coalitions, and constraints from local information are examined.
Quarto
Swiss army knife for hand-whittling scientific reports, slides and blogs
Wherein the Pandoc-based publishing system is presented, and its rendering of Jupyter notebook sources with functioning citations, multi-language computations, and varied outputs (HTML, PDF, Word) is recorded.
Reinforcement learning
Wherein a historical machine is described, in which tic‑tac‑toe play is produced by matchboxes of beads, outcomes are used to adjust bead counts, and learning is recast as reward‑driven decision making in MDPs.
Particle variational message passing
Graphical inference using empirical distribution estimates
Wherein an importance-sampling particle analogue of variational message passing is described, and updates are effected by importance-weighted empirical CDF ensembles rather than by parametric factor summaries.
Web browser automation
Industrialising the ancient handicrafts of serfing the web and tilling the clickfarm etc
Wherein browser drudgery is delegated to scripts, Playwright is noted as a testing framework repurposed for full browser automation using headless instances, and TabFS is described as mapping tabs to folders.
q-exponential process
Wherein a stochastic q-exponential process is proposed as a probabilistic analog of L_q regularization, is linked to Besov spaces, and is specified via elliptic-contour constructions enabling tractable prediction
Recurrent / convolutional / state-space
Translating between means of approximating time series dynamics
Wherein the relationships between recurrent, convolutional, and state‑space models are examined, and the S4 method’s structured diagonalization enabling efficient long‑range sequence handling is described.
Prepping for the end of (my access to) the (industrialised) world
Wherein communal measures for post‑industrial survival are catalogued, and a basic go‑bag plus neighbour‑led mutual aid hubs are presented as concrete, practicable first steps to be assembled.
Singapore
Wherein Singapore is described as an island city-state in which a 1977 action heroine, Cleopatra Wong, is noted for performing daring stunts and for anecdotally safeguarding regional currency stability.
Python
Syntactic saccharine for compiled code
Wherein Python is described as a practical lingua franca, being used to stitch together C, Fortran and R libraries, and to run within browsers via Pyodide, for varied scientific tooling.
Knowledge geometry
Wherein the adjacency of disciplines is considered, and embeddings of papers into hyperbolic metric spaces and citation networks are examined as models for the growth and gaps of collective knowledge.
AI search
Retrieval-augmented generation for the working schlub
Wherein AI search over a bounded corpus is described, and vector embeddings such as SPECTER2 are trialled with ChromaDB storage to enable retrieval-augmented similarity and similar-post discovery.
Machine learning for inverting partial differential equations
Wherein tomography through partial differential equations is undertaken by learned solution operators and neural priors, and computational inversion is accelerated by data-driven surrogate models.
Graphical model / machine learning decoder ring
I’m thinking something through for myself. Details are absent right now. Twin to Causality+ML, perhaps.
Conflict theoretic models
Wherein the study is presented as an account of group coordination, and rhetoric is dissected as a competitive power game, while elite consensus and norms enforcement via altruistic punishment are examined.
QR codes
Wherein QR codes are described, and Python libraries such as PyQRCode and qrcode are indicated for local generation, while URL shorteners like bit.ly are recommended for click-through tracking.
Variational Bayes Neural Nets and Graphical Models
Wherein variational Bayes for neural nets and graphical models is considered, and scalable approximate inference for large data, low‑rank parameterisations, and inverse‑problem posteriors is examined.
Economics of automation,
Wherein the reordering of human tasks by mechanized intelligence is described, and its effects on job-task composition and wage distribution—especially within service-sector labor—are presented, and the shifting share of manufacturing is noted.
PDF, Portable Document Format
Wherein the quirks of the Portable Document Format are catalogued, including commands to downgrade PDFs to v1.4, tools to extract embedded tables, and methods to reduce file bloat.
Contemporary rationalists
Cloth mother/wire mother, but for cognition
Discussion of an open community that specifically aims to engage with arguments despite edginess. Only ever one click away from finding dialogues with people with…
Databases (for scientists)
Stashing experimental data for later use
Wherein scientific database choices are surveyed, and a preference for serverless, file-backed storage for gigabyte-scale experiment data (HDF5, SQLite, DuckDB) is articulated, and concurrent-write tradeoffs are catalogued.
DIY social networks and social-groupware
V. cypherpunk! much 733+! Underground!
Wherein methods for self-hosted and decentralised community platforms are surveyed, and the problem of organising IRL events via shared social calendars is noted.
Indieweb, small web, cozy web
Wherein the reader is introduced to a hand-crafted internet of small, autonomous sites, and the practice of POSSE alongside protocols such as webmentions and IndieAuth is described.
Secure chat systems
Optimizing back channel interjections into other people’s meetings
Wherein are surveyed the tradeoffs of modern end-to-end messengers, the leakage of metadata and difficulty of exporting archives, and the advisability of face-to-face parleys within a Faraday cage.
Human domestication
Wherein humans are proposed to be domesticated relative to their institutions, and a mathematization is sketched in which individual agency is assessed by causal influence over systems, with traces of operant conditioning considered.
Editing images using a GUI
Chinks in my armour of learned Photoshop helplessness
Wherein the available graphical editors are surveyed, and their extensibility by user‑written plugins in Python, C#, or Java for multidimensional scientific imagery is detailed.
Money, Australian-style
Wherein Australian personal finance is surveyed in the manner of a practical ledger, and the legacy of a 19th‑century gold‑commodity economy is recorded as shaping tax, real estate and investment practice.
Neural denoising diffusion models with non-Gaussian distributions
Wherein neural denoising diffusion models are examined on non‑Euclidean manifolds, their applicability to partial differential equations and non‑Gaussian spaces is outlined, and a recent discovery by the author is noted.
Git tricks
Wherein a pragmatic workaround is described for keeping a repository’s .git metadata off cloud‑synced folders by using a bare repository plus a separate .git file, so normal git operations are preserved.
Pathwise solutions of stochastic differential equations
Wherein stochastic dynamics are treated pathwise by smoothing Brownian trajectories and solving the resulting ODEs, so that Wong–Zakai approximations are shown to converge to Stratonovich limits.
Scalable vector graphics
Wherein the SVG format is presented as a hub for vector interchange, and its interoperability with PDF and with conversion utilities such as Inkscape, dvisvgm, and web‑based generators is detailed.
Generative AI workflows and hacks 2025
Wherein practical notes are presented on installing local LLMs via Ollama and LM Studio, command-line pipelines using files-to-prompt → llm are detailed, and model-cost implications are recorded.
Text data processing
Wherein the peculiar arts of text data processing are surveyed, and command-line contrivances for CSV, JSON and LLM‑assisted scripting are evoked, including jq, xsv, llm and ttok.
Transformer networks
Wherein the architecture is described as employing self‑attention and positional encodings, and scaling laws are invoked to permit extremely large, trainable language models.
Won’t somebody think of the children?
The bottom rung of the status ladder
Wherein the rhetorical elevation of children is examined, and the invocation “for the children” is shown to be used to justify policies from climate planning to AI regulation, irrespective of measured effect.
Multimodal AI
Wherein a shared semantic space between text, image, and audio is constructed by contrastive pretraining and is used to condition diffusion generation via CLIP embeddings, enabling cross‑modal retrieval and control.
Differentiable learning of collective automata
Wherein classical cellular automata are relaxed to continuous states and are trained by gradient-based learning using local Sobel-filtered gradients, and emergent patterns are framed as dynamical attractors.
Plotly
Wherein Plotly is presented as a plotting system whose browser‑based interactivity is described and whose non‑browser image export via orca or kaleido and multi‑language wrappers are noted
Designing social movements
Synthetic biology for utopian egregores
Wherein the mechanics of recruitment and institutional hygiene are examined, and Schelling-style coordination models are invoked to explain how movements self-replicate and police membership.
Email blogs and newsletters
Wherein email incarnations of blogs are surveyed, platform trade‑offs are catalogued, and Substack’s lack of an API, Ghost’s open‑source hosting option, and Buttondown’s privacy‑first defaults are reported.
Schelling-Goodhart coordination problems
Wherein coordination failures on shared belief maps are examined, and the use of Schelling points to render questions unaskable in culture wars and taboos is described, with attention to misrepresentation and preference cascades.
Science communication of ML research
Wherein methods for relaying AI and ML findings to publics and peers are examined, with attention to safety and economics, and a noted tendency to slip into debate rather than sustained canvassing.
Indonesia
Wherein the emergence of online shopping in Indonesia is documented, difficulties in international shipping to Australia are noted, customary musyawarah mufakat practices are described, and past state violence is evoked.
Feedback system identification, not necessarily linear
Wherein nonlinear feedback systems are addressed via state filtering and variational methods, and recursive estimation of fixed parameters is shown to be achievable using simulation-based or RKHS-instrumental techniques
Melbourne / Naarm
Australia’s counterculture capital
Wherein the city’s dual naming as Naarm is noted and it is observed that the bush lies farther from town than in Sydney, so native superb wrens are encountered only rarely.
uv, the python package manager
A hipster python package manager with handy features for my workflow
Wherein the Python package manager is described, and its deep PyTorch GPU-specific support is noted; it is written in Rust, is claimed to replace many tools, and is presented as version-managing.
Neural denoising diffusion models with non-Gaussian distributions
Wherein diffusion models for categorical and Poisson-like data are treated, and physics‑inspired embeddings in N+D space with tunable D are described to connect Poisson flow and diffusion approaches
Privacy while web browsing
Browsing the internet without giving corporations my personal information for free
Wherein the browser is outfitted with extensions and compartmentalized containers to limit fingerprinting and third‑party trackers, and HTTPS‑only connections are advised for sensitive exchanges.
Human enthalpy
Nudges, changing things, UX for life
Wherein small amounts of activation energy are described as producing large shifts in behaviour, the region‑beta paradox is invoked to explain threshold effects, and personal anecdotes and policy examples are cited.
Machine learning for partial differential equations using diffusion models
Neural physics for neural physics
Wherein diffusion models are employed to learn PDEs, Poisson flow generative models are noted, manifold-valued solution spaces are considered for non-Euclidean PDE learning, and PFGM++ code is cited.
Voice transcriptions and speech recognition
Wherein speech is converted to text, real-time dictation and batch transcription are surveyed, with Whisper for recordings and Talon, Serenade and voice-programming workflows noted, and background music issues described.
Neural denoising diffusion models of language
Wherein a class of neural diffusion techniques is adapted to discrete text tokens, and a non‑Gaussian, discrete diffusion process is described as an alternative to autoregressive generation in low‑data regimes
Generative art with language+diffusion models
also some autoregressive models
Generative art using modern diffusion-backed image generators. The name-brand models are DALL-E 2, Stable Diffusion, Midjourney etc., which are diffusion models for image…
Free images
Wherein sources of freely usable images are catalogued, and the Rijksmuseum’s high‑resolution scans of 17th–18th‑century Dutch engravings are noted as accessible via translated search queries.
Neural nets that do symbolic mathematics, logic and other reasoning tasks
Wherein models are described as being taught to double-check their own reasoning by appending the token Wait at test time, and a curated s1K dataset plus budget-forcing are reported to boost competition-math performance.
Multi-agent self
Wherein the self is portrayed as divisible agents, and psychotherapy’s Internal Family Systems alongside shard theory are offered as compositional frameworks for explaining intrapersonal roles and conflicts.
Syncthing
A good, easy, peer-to-peer, free Dropbox-replacement system for the modern paranoid age
Wherein Syncthing is presented as a peer-to-peer file synchroniser without a central server, using TLS encryption, and is used to synchronise music-production folders across studio, backup and gig machines.
Python packaging, environment and dependency management
“import antigravity” raises an exception if you want CUDA antigravity
conda pip+venv poetry uv.
Pip-like python environment management
The mainline family of python package managers
pip
is the default Python package installer. It is the most widely used and most widely supported package manager for Python, especially if you include the many derivative…
Multi agent causality
Game theory and decision theory for lots of interacting agents
Wherein agents' decisions are modeled via a mechanised multi-agent influence diagram (MMAID) extending causal DAGs to iterated games, and the problem of commitment races is examined for AI safety.
Neural PDE operator learning on domains with interesting geometry
Can ML solve PDEs on non-Euclidean domains?
Wherein implicit diffeomorphisms are learned to map Euclidean grids onto irregular manifolds, and neural operators are adapted to mask and represent partial differential equations on such domains.
Neural PDE operator learning
Especially forward operators. Image-to-image regression, where the images encode a physical process.
Wherein the Fourier transform is employed to learn PDE forward-propagators as resolution-agnostic operators, and the Burgers’ equation is used as a concrete example of one-step propagation learned from simulation timesteps.
The AI tech soap opera
Wherein the corporate choreography of AI is described, with Microsoft’s partnership with OpenAI and leaked executive correspondence recounted as elements shaping the industry’s unfolding narrative.
Feed readers
A standard for user-driven, open news
Wherein the revival of subscription feeds is described, Feedly’s ranking algorithm is noted as surfacing academic specialist blogs, and self‑hosted readers such as FreshRSS are outlined for private use.
Learning with conservation laws, invariances and symmetries
Wherein conservation laws are examined in the context of nonparametric and overparameterized neural models, and methods such as Lagrangian networks and Hamiltonian Monte Carlo are described
Causality, agency, decisions
Exotic decision theories, Newcomb’s boxes…
Wherein mechanised causal graphs are used to extend notions of causality to agents whose choices feed back into outcomes, and decision paradoxes such as Newcomb-style problems are examined.
Aligning AI systems
Practical approaches to domesticating wild models. RLHF, Constitutional AI, etc
Wherein practical methods such as reinforcement learning from human feedback and mechanistic interpretability are surveyed, computational morality is noted, and the inherent fuzziness of alignment is acknowledged.
Conda-like Python environments
A parallel python package system to pip
Wherein the conda model is described, and it is noted that the distribution and the package manager are separately licensed, so commercial licensing risk for Anaconda distributions is made explicit.
Poetry, the python package manager
A python package manager that I briefly used
Wherein poetry is presented as a single-file project manager, is noted to create and manage local virtual environments, is described as offering its own dependency resolver, and is taxed by CUDA/GPU installs.
Leadership
The default pseudo-content for undirected MBA research
Wherein the efficacy of leaders is examined, and it is shown that monarchs' cognitive ability—instrumented by royal inbreeding—is linked to state performance in reigns where power is unchecked.
Reading ebooks
Wherein desktop handling of e‑books is presented, options are catalogued, Calibre’s ISBN‑oriented management is noted, and Thorium’s EPUB‑3 support and accessibility for LCP‑protected files are described.
Anti-TESCREALists
Post-rationalism for non-post-rationalists
Wherein the coinage TESCREAL is presented as an acronym for Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism and Longtermism, and its othering function is described.
Computational complexity of Bayesian inference
Wherein the difficulty of Bayesian inference is examined, and the contrast between sampling-based posterior approximation and exact calculation on graphical models and on arbitrary measure spaces is presented.
Neural PDE operator learning using transformers
Wherein transformer-based operator learning for PDEs is examined, experimental integration with existing notebooks and benchmark comparisons are proposed, and diffusion-style alternative models are contemplated.
Adaptive design of experiments
I am not going to call it ‘Bayesian optimization’, but that is what everyone else does
Wherein adaptive experiment design is presented as surrogate-based optimisation for expensive noisy black-box functions, Gaussian process surrogates and acquisition functions are surveyed, and Ax is noted for deployment.
Causal Bayesian networks via probability trees
Wherein causal relations are expounded through sequential probability trees, interventions are enacted by fixing branches and temporal ordering is shown to afford identifiability in lieu of do-calculus.
Travel hacks
Wherein practical travel hacks are catalogued, and techniques for compact shirt folding and procuring travel eSIMs are described, with notes on comparing Airbnb internet speeds using browser extensions.
Inference from disorder
Wherein inference is drawn from lack of structure, and the arrow of time is inferred from asymmetries in noise‑outsourcing conditions, while methods for causal ordering and entropy are examined.
Artificial agency
Wherein the question of agency is examined via causality-based models, the emergence of self in machines is contemplated, and the possibility that the human is not the agent in collaborations is considered.
Neural codecs and compression algorithms
Wherein neural networks are employed to compress images, audio, and video for transmission, a 90× audio codec is presented, and images are resampled into flexible-length 1D token sequences via vector embeddings
Machine learning for partial differential equations via flows
Wherein a normalising‑flow approach is applied to PDE learning in function space, and Gaussian process priors with Matérn kernels are employed so that function‑space ODEs map Gaussian noise to target solutions.
Causal inference on DAGs
Wherein graphs are invoked to decide when effects are identifiable from observational data via d-separation and do-calculus, and criteria such as back-door and front-door adjustment are described.
Scaling laws for very large neural nets
Theory of trading-off budgets for compute size and data
Wherein the behaviour of neural networks is examined as parameters approach billions, and an observational scaling method using about 100 public models is proposed, in which a low-dimensional capability space and sigmoidal compute trends are identified to predict novel capabilities
The simplest thing
Wherein the problem of determining the least effortful course of action is examined, and the concrete test case of rewriting software is presented to show who is made to bear irreducible complexity.
Bayesian and causal inference by foundation models
Wherein transformers are examined as generalized inference machines and is argued that in‑context learning is modeled as Bayesian conditioning, with twisted sequential Monte Carlo methods proposed for sampling.
Data summarization
a.k.a Data distillation. sketching
Wherein methods for compressing datasets into weighted subsets such as coresets or representative subsets are surveyed, and connections to Bayes duality, sketching, and influence-based selection are outlined.
Extraversion
Wherein the nature of extraversion is recounted, its linkage to dopamine-driven reward learning is described, and the author’s personal stake as a high-scoring extravert is disclosed.
Causal inference under feedback
Wherein feedback loops are examined and it is shown that effective control can erase observational correlation, and extensions to continuous spatial fields and formal treatments by Central European causality researchers are surveyed.
Causal inference in highly parameterized ML
Wherein the application of causal graphs to neural nets and other nonparametric models is considered, dataset shift and invariance principles are examined, and benchmarking platforms for discovery are surveyed.
The Netherlands
Wherein the origins of several capitalist practices are traced to the Low Countries, and the first major financial bubbles are observed, while colonial ties to Indonesia are noted and Rijksmuseum holdings are consulted.
Peter Piper’s Practical Python Pickling Patterns for Parallel Processing
Wherein Python pickling for parallel processes is described, cloudpickle is shown to serialize lambdas for distributed execution, and joblib plus compression methods are advised to reduce inter-process transfer overhead
Localhost dev server
Wherein local HTML pages are served for testing, and a simple python3 -m http.server bound to 127.0.0.1 is used, while considerations of TLS, Caddy auto‑SSL, tunnelling, and live‑reload are catalogued.
Unix commands I need often
but which are tedious to work anew out each time
Wherein modern replacements for classic Unix utilities such as ripgrep, fd and bat are catalogued, and techniques for file‑watching, resource monitoring and cross‑platform prompts are described.
Majority illusions and filter bubbles
Wherein the reader is apprised that, in homophilic social networks, a few highly connected hubs cause rare opinions to be made to appear common in local samples, thereby biasing prevalence estimates.
Email clients on linux
Unix invented email, and kinda left it there for the next 50 years
Wherein the available Linux email clients are surveyed, and the desirability of local offline clients with calendar and contact integration is contrasted with their prevalent usability and sync shortcomings.
Urbanism
Wherein urban order is surveyed, and debates from smart‑city surveillance to charter‑city experiments are presented, while city scaling laws linking population to energy use are examined.
Technological determinism
Technium, amistics, the robo-omega-point etc
Wherein technological change is presented as steered by tools already forged, and energy flows and urban-scaling data are adduced as concrete measures, while a personal notebook is reported lost.
Western Australia
A big state with modest dreams
Wherein the vast state is described; its landmass of 2,527,000 km² is noted and the capital, Perth, of about 2.14 million souls, is observed to favour secluded suburban life.
Collaborative intelligence
Wherein the partnership of human judgment and machine pattern‑finding is examined, and the rise of reinforcement learning from human feedback as a method of alignment is described, alongside risks of reverse‑centaurs.
Hydra for configuring and tracking machine learning experiments
Wherein Hydra is described as a generic configuration system for ML, a dedicated output directory per run is created to record the exact parameters and artifacts, and optional plugins for parallel parameter sweeps are supported.
Calendars and contacts GUIs for Linux
Wherein various Linux GUIs for calendars and contacts are surveyed, and CardDAV/CalDAV interoperability is reported, noting iCloud quirks and Flatpak-era GNOME/Evolution configuration pitfalls.
Gibbs posteriors
Bayes-like inference with losses instead of likelihoods
Wherein the posterior is constructed from a loss in place of a likelihood, and a tempering learning rate ω is introduced to control the influence of empirical risk, thereby linking inference to energy‑based models.
Organising a music collection
Also guessing missing metadata
Wherein music files are catalogued and analysed for key and BPM, associated artwork is managed en masse, and tools such as beets and MusicBrainz Picard are surveyed for metadata correction and bulk renaming.
Southeast Asia
Wherein the author’s long residence across Australia, Indonesia and Thailand is recorded, and a 17th‑century Swiss soldier’s itinerary in Dutch Formosa is noted as a historical curiosity, while the contention that Australia is excluded is registered.
Advice on advice
Wherein the perils of generic counsel are expounded, and a hierarchy of advisory skill — from reaction to causal modeling — is set forth, while the difficulty of personal calibration is noted.
Visual Studio Code for prose
Wherein VS Code is treated as a prose workbench, and its support for Markdown mathematics and a variety of spell‑checking extensions (cspell, SpellRight, LTEX) is catalogued for editing workflows.
Chromium browsers
Wherein the dominance of a single engine is noted, and alternative Chromium-based browsers are presented with concrete particulars such as Vivaldi’s built-in mail and calendar and Brave’s Linux emoji quirks.
DJing
On encouraging people to listen to your living room playlist by demanding that they pay for it
Wherein the craft of DJing is delineated as a global soundtrack of both capital and its opposition, and the practical note is given that Pioneer-compatible USB keys are carried to gigs.
Continuing to do obviously complicated things in a naïve way that doesn’t work
Also, not researching alternatives
Wherein a decade of sclerotic cohousing meetings is endured before governance literature is consulted, and the persistence of unresearched amateur strategies is described in measured, passive terms.
Gaussian Ensemble Belief Propagation
Wherein a novel Gaussian ensemble belief propagation is presented, its ensemble updates are used to update million‑pixel fluid‑simulation emulators, and it is noted that the work is accepted at ICLR 2025.
Fediverse
Wherein the fediverse is described as a volunteer-run, ActivityPub‑based federation of services such as Mastodon, Lemmy and PeerTube, and is hosted across myriad independent servers maintained by volunteers.
Python CLIs
Putting the “argument” in “command-line argument”
Wherein Python command-line interfaces are surveyed and their roles in configuring ML experiments with Hydra or pyrallis are noted, while multiple parser libraries and dependency overlaps are catalogued.
Misinformation, disinformation etc
Trolls, bots, lulz, infowars and other moods of the modern networked ape
Discussion of terrorism and hate speech
Attention management tips for web browsing
Wherein attention on the web is corralled through browser tweaks, extensions, and site blockers, simple tracking tools are described to record usage, and intrusive Google sign‑in popups are shown to be blockable to prevent modal interruptions.
Voice fakes
Wherein voice fakery is surveyed as a technical practice, and style-transfer methods are described as enabling cloning from seconds of speech but are noted to demand multi-gigabyte GPUs and tedious training.
Windows Subsystem for Linux
Wherein a miniature Linux distribution is reported to be installed and run within Windows via WSL, and the convenience of executing native Linux setup commands and tools from a Windows terminal is noted.
Distilling Neural Nets
Wherein a procedure is related in which a larger teacher model is employed to synthesize training examples, and a smaller student model is thereby instructed to emulate its probabilistic outputs for deployment.
submitit
A less horrible way to parallelize on HPC clusters
Wherein submitit is presented as a job-submission wrapper for SLURM, and the executor is employed to serialize and reload job handles to disk so queued Python tasks can be resumed from a later session.
Diffusion of innovations
Wherein the geographic concentration of economically impactful technologies is recorded, the slow fifty-year dispersion of related jobs across regions is described, and an initial high-skill bias is observed to decline.
Operationalising the bitter lessons in compute and cleverness
Amortizing the cost of being smart
Wherein it is argued that massive training compute is justified because inference is cheaply amortised, enabling foundation models trained once to serve many tasks, and token-level economics and compute overhangs are examined
VS Code online and networked
Wherein remote usage of VS Code is treated, and a procedure for attaching a Python debugger via debugpy on a forwarded port is outlined, with SSH tunnels and tmux persistence being noted.
Making good intertemporal decisions
Wherein the intrapersonal collective‑action problem of conflicting selves across time is described, and interventions like role‑playing an advising counsellor and contractual self‑commitments are examined.
Sparse regression
Wherein penalised regression with LASSO-style absolute-coefficient penalties is presented, and sparse models are attained by tuning a penalty parameter under quadratic or generalised losses for predictor selection.
Regression with functional data
Wherein curves are regressed against predictors by being represented in functional bases, with derivatives and Karhunen–Loève components being used to capture dynamics and to permit warping-based alignment.
Dynamics of recommender systems at societal scale
Variational approximations to high modernism
Wherein societies are shown to be steered by recommender systems into Matthew effects that amplify popularity, while participatory audits such as Tournesol are convened to measure and surface public‑interest content
Goodhart’s Law
Wherein the adage is outlined and four failure modes—regressional, extremal, causal, adversarial—are described, and the heightened importance under strong AI-driven optimisation is noted.
Ageing
Both descriptive and prescriptive
Wherein the pursuit of extended life is examined and the prospect of rejuvenation by diluting aged blood with saline and albumin is presented, while quackery is warned against and biomarkers are queried.
The blogosphere
Now with added newsletter-o-sphere
Nascent producers and consumers of online punditry
Blame engineering
On institutions for siphoning responsibility
Wherein are examined the mechanisms by which blame is engineered within bureaucracies and markets, and the diffusion of responsibility across institutions and persons is delineated.
GP inducing features
Wherein inter-domain inducing features are presented as linear operators yielding RKHS basis functions φ_m via integral transforms, predictive distributions are obtained by marginalising u, and parameters are learned by KL minimisation.
Artificial intelligence without (necessarily) using computers
Are corporations artificial intelligences? How about states? How about my local bowling team?
Wherein it is presented that the Industrial Revolution is to be regarded as a past Singularity, and that markets and bureaucracies are depicted as distributed information-processing systems whose reach is expanding.
Dunning-Kruger theory of mind
Complicated questions where we would all prefer simple answers
Making macOS behave itself
Things I have to do to keep my laptop running so I can google how to fix other things
Wherein various means of making macOS comport are recounted, with instructions for preventing system alert beeps from sounding on unintended Bluetooth or HDMI devices, and for toggling notarization checks to reduce launch delays.
AI Safety
Getting ready for the grown-ups to arrive
Wherein a survey of AI safety is presented, and a living repository of 777 categorized risks assembled from 43 taxonomies is described, with taxonomies of causal factors and domains outlined.
Governance of and by AI
Wherein the governance of and by AI are examined in a measured survey, and the rise of national AI safety institutes and legislative briefings is noted as a concrete locus for regulatory and alignment experiments.
Teaching mathematics and especially statistics
Wherein curricula and resources for introducing statistics are surveyed, with emphasis on intuition‑building, Bayesian and computational approaches, and practical tools such as R and probabilistic spreadsheets.
Probability
Wherein probability is treated as a calculus of models for observed regularities, and a conditional-probability foundation via Rényi axioms is presented as guiding Bayesian belief updates.
Transformer networks as recurrent or state-space models
Wherein recurrent and state‑space variants of transformers are surveyed, and models such as RWKV and Mamba are noted to enable linear‑time scaling and million‑token contexts without relying on classic attention.
Morality and computational constraints
It is as if we knew what we were doing
Wherein connections between moral theory and computation are examined, with questions posed about whether reward signals correspond to pleasure or pain and whether RL practices are considered to entail workplace safety obligations.
Learning Gaussian processes which map functions to functions
Wherein operators between function spaces are treated as Gaussian measures on Hilbert spaces, and the subtleties of infinite-dimensional equivalence and mutual singularity (Feldman–Hájek) are examined.
AI Alignment Fast-Track Course
Scattered notes from the floor
Wherein failure modes and mitigation techniques are examined, RLHF and scalable oversight are surveyed, adversarial attacks and convergent instrumental goals are discussed, and Ajeya Cotra’s taxonomy is introduced.
Prefigurative politics
Anticipatory politics, building a new world in the shell of the old etc,
Wherein the merits and costs of embodying one’s ideals in movement practice are considered, and an Australian racial-equity organisation is noted to be staffed solely by Anglo‑Australians.
Trading securities, hedging and portfolio design, practical
Shares, cryptocurrency, options and derivatives for babies
Wherein portfolio hedging, risk‑parity construction and algorithmic trading infrastructure are surveyed, and an R package for rapid risk‑parity design is noted as a practical tool.
Advice calibration
Wherein the difficulty of tailoring population-level counsel to individuals is examined, with attention to interaction effects, self-selected readerships that distort external validity, and an eating-disorder example.
Personal finance
Various notes on the concept of being minimally burdened by money problems
Wherein joint optimisation of money and life satisfaction is proposed, Australian sharehouse accounting woes are examined, and virtual split-payment tools such as Beem are surveyed, with retirement investing and monetisation of internet personas briefly addressed.
Alt finance
Wherein a catalogue of contrarian financial analysts is presented, including newsletters and essays, and practical tips on money management alongside speculative commentary are offered in lucid, restrained prose
Statistical mechanics of statistics and NNs
Wherein phase transitions in statistical inference, notably a detectability threshold for community detection, are described and connections to neural network loss landscapes are surveyed.
Digital forensics
OSINT, deep fakes, and the provenance of information
Wherein the provenance of information is examined as it is complicated by deepfakes, and methods such as satellite imagery analysis and open‑source investigation toolkits are indicated for forensic corroboration.
The production of bullshit
Wherein the ubiquity of proxies and practices such as tufta, Goodhartian gaming, and KPI-driven bureaucracy are examined as mechanisms by which seeming progress is produced.
Actually-existing capitalism
Wherein a compendium of satirical links on gig‑economy fraud, algorithmic arbitrage, corporate climate calculus, VR cows, and the commodification of human contact is presented.
Machine learning for Bushfires
Wherein spatiotemporal deep learning is applied to satellite and sensor data to analyse and predict bushfire spread, and foundation-model approaches are examined for geospatial tasks.
Data sovereignty
Wherein the notion of data sovereignty is set as ownership and control of personal and communal records, and data unions are presented as pooled streams governed by smart contracts that apportion revenue to contributors.
Implementing neural nets
The internet is full of guides to training neural nets. Here are some selected highlights.
Single-site web browsers
Wherein a method is described whereby individual websites are instantiated as standalone desktop applications, their cookie and storage contexts kept isolated as on macOS Safari Save to Dock, enabling separate logins.
Game theory
Wherein the Prisoner’s Dilemma is presented, and the tension between dominant defection and Pareto‑efficient cooperation is depicted alongside notes on iterated and stochastic games and the Shapley value
Verifiable information, identities on the internet
Now that it is cheap to fabricate history, how can we trust anything?
Wherein cryptographically signed provenance and adversarial digital forensics are proposed as remedies for cheaply fabricated histories, and the problem of verifying online identities is framed.
Learning as compression
Wherein learning is treated as compression; the reduction of model description length by counting parameters and invoking Occam’s razor is examined, and psychometrics is proposed as a case study.
Agent-based models
Wherein a bottom-up method is presented, individual agents and their interactions being modeled to probe emergent behaviours, with applications ranging from markets to nation-scale epidemic simulations.
User interface software
Wherein a compendium of minimal UI libraries is presented, and immediate-mode browser and Electron bindings and an emphasis on WebAssembly first-class support for cross-platform embedded graphics are noted.
Sexual ethics and institutions
Wherein modern Western experiments in relationships are surveyed, terminology such as polycules and metamours is introduced, and sexual ethics and institutional politics are traced through recent and historical cases.
Visualising geospatial data
Geographic information systems, or, as we in the trade refer to them, “maps”
Wherein the peculiar pitfalls of cartographic data visualisation are considered, and the practice of encoding geodata as GeoJSON for browser maps using OpenStreetMap tiles is described.
Machine learning for biology
Alphafold, connectomics, and other applications
Wherein approaches for expediting cellular experiments and enhancing dataset quality are outlined, with emphasis on shortening experiment cycles and collecting higher-fidelity cell-level measurements.
Scientific machine learning
Machine learning for physical sciences, SciML
Wherein machine learning is employed to infer best-possible stochastic models of physical systems, ocean tracer dynamics and conservation laws are invoked as constraints, and surrogate PDE-informed networks are considered.
Generative AI workflows and hacks 2024
Wherein ephemeral notes on generative AI workflows are presented, and practical notes on self‑hosted chat GUIs, their endpoint‑switching limits and onerous install steps are recorded, and standards and local runtimes are catalogued.
Embodiment
On the hypothesis that it is easier to understand walking if you have legs
Wherein the necessity of a body to mind is examined, distinctions of strong and weak embodiment are delineated, and human thought is reported to operate at roughly 10 bits per second.
Monetizing my music
Wherein the economics of circulation are described, streaming royalties and playlist payola are examined, and subscription models, Bandcamp, and the crypto platform Audius are noted.
Timeless works of art
Aesthetics as a conditional average treatment effect
Wherein the persistence of artworks is interrogated and the encroachment of algorithmic, hyperproduced popular forms—AI-generated music and mass-produced kitsch—is described as reshaping reception contexts.
Learning of manifolds
Also topological data analysis; other hip names to follow
Wherein manifold-based dimensionality reduction is set forth, and the application to genomic search is adduced, wherein compressive manifold storage based on fractal dimension and metric entropy is employed.
Home automation
Wherein a Raspberry Pi is pressed into service as a home automation hub, Zigbee and Matter are compared, and cloud‑dependent devices are noted to leak telemetry over Wi‑Fi.
(Geo)spatial data sets
Wherein thirty years of satellite imagery totalling over twenty petabytes are surveyed, cloud-based APIs and catalogs for rainfall, tomography, biodiversity and POI layers are noted for remote processing and discovery.
Our microbiome
Wherein faecal microbiota transplantation is reported as being practised in Sydney, commercial supplies of medical stool are noted, and the body is portrayed as an ecosystem whose cells and genes are largely nonhuman.
State capacity
Especially applied to western democracies, especially Australia
A mess of links about the state and its capacity to do things, incorporating institutions, social licence and the like.
Statistical learning theory
Wherein finite-sample generalisation bounds are surveyed, and Rademacher complexity, VC dimension, estimator stability, and extensions to non‑i.i.d. data are documented.
The Score Function Estimator
a.k.a. REINFORCE; a gradient estimator for expectations
Wherein the log‑derivative estimator is presented as a generic gradient method, a PyTorch demonstration on categorical distributions is given, and its high Monte Carlo variance under small batches is noted.
Constructivist rationalism
Wherein Hayek’s constructivist rationalism is examined, a contrast with critical rationalism is outlined, and a boiling‑point experiment with a Bunsen burner is invoked to show how complex practices are allowed to emerge.
tmux
Wherein terminal sessions are managed and are persisted across logins; mouse-driven scrolling and clipboard handling are treated as separate behaviors, and plugins together with iTerm2 integration are noted.
Bayes neural nets via subsetting weights
Wherein the practice of treating a subset of neural network weights as random variates while others are held deterministic is examined, and sequential Monte Carlo training is described.
Quantum computing for ML
Wherein the prospects for quantum-accelerated machine learning are examined, and concrete conditions for provable speedups—data access models and noise limits—are delineated.
Particle filters
Wherein a population of samples is updated by nested sequential Monte Carlo steps to track time-evolving states, and system parameters are learned via particle marginal Metropolis–Hastings.
Thermodynamics of life
Wherein the statistical mechanics of organisms is surveyed, and recent work linking non‑equilibrium fluctuation theorems, cellular information processing, and metabolic theories of life is presented.
Linux-compatible laptops
I love linux but I hate googling for “wifi hangs after suspend”
Wherein an inventory of Linux-compatible laptops is presented and their Thunderbolt 3 external‑GPU support and battery‑management quirks are examined for practical maintainability.
Australian AI Safety Forum Sydney 2024
Scattered notes from the floor
Wherein a Sydney forum is convened in November 2024 to present technical and governance papers, a nuclear‑safety‑case analogy for AI is advanced, red‑teaming difficulties are reported, and talks are archived online.
Machine learning for climate systems
Wherein machine learning for climate systems is addressed, and satellite retrievals from Sentinel‑3 are noted to detect methane plumes of about ten tonnes per hour under suitable conditions.
Disseminating science
Journals and preprint servers etc
Wherein the mechanisms of scholarly dissemination are surveyed, and it is noted that pirate shadow libraries such as Sci‑Hub are relied upon in contexts like Indonesian academia to access paywalled research.
Tabular data processing in Python
Wherein Python approaches to tabular processing are surveyed, and a contrast is drawn between pandas' row‑labelled DataFrames and Polars' Arrow‑backed, index‑free, Rust‑implemented tables for faster queries.
Website cheat codes
CSS, SCSS, SASS, HTML, UX, Web 2.0, RFC, Yeah you know me
Wherein practical web conveniences are catalogued and a method for generating multi-resolution favicons via Real Favicon Generator is noted, and various dev and encoding tools are surveyed.
Criminal justice
The ethics and efficacy of policing and punishment
Wherein the turn toward probabilistic law and automated statistics is examined, and prisons are contrasted with behavioural and social‑norm interventions as means by which reoffending is reduced.
Asynchronous Python
It can’t be premature optimisation if it took 20 years to start
Wherein asynchronous Python ecosystems are surveyed and uvloop performance alternative is covered, Trio and asyncio are contrasted, and common libraries such as HTTPX, aiohttp and pyzmq are noted.
TikZ/PGFplots etx
Wherein TikZ is described, and diagrams are rendered in Quarto via a custom extension, while PGF/TikZ compatibility and manual layout control are contrasted with automatic layout tools.
The robot regency
How long is it safe to have our learning left unsupervised?
Wherein a wealthy parent’s child is depicted being provided with personalised assistants and a whole consultancy is constructed around her due to developmental delay, and questions of autonomy and dignity are posed
Jax
Wherein JAX is presented as a numerical library, whose NumPy programs are compiled via XLA to run on GPUs and TPUs, and whose composable, pure-function transformations are offered as grad, jit and vmap.
Differentiable PDE solvers
Wherein differentiable PDE solvers are presented as frameworks that expose discrete adjoint gradients for machine‑learning integration, and specific tools like PhiFlow and dolfin‑adjoint are cited.
Machine learning for partial differential equations
Wherein neural surrogates for PDE solvers are described, operator learning of time‑stepping propagators is emphasized, grid‑free representations for spatiotemporal fields are noted, and inverse problems are considered.
Canalization and plasticity in human brains
Updating fast-system priors; therapy as described by mathematicians
Wherein grey-matter reductions across pregnancy are described, canalization is set against neuroplasticity and puberty-like plasticity induced by psychedelics is noted, and the notion of annealing is considered.
The Jupyter Cinematic Universe
Wherein the browser‑based notebook is described as a network‑friendly remote execution interface, is built atop kernels and .ipynb files, and is contrasted with emergent, slimmer Python‑specific alternatives.
VS Code / VS Codium
Egg-laying wool milk code editor
Wherein the community-built Codium is presented as a telemetry-free distribution of VS Code, its exclusion of Microsoft telemetry is noted, and package-repo installation steps for Linux, macOS and Windows are sketched.
Learning stuff
Shoving knowledge into my brain
Wherein AI-assisted tutoring, spaced repetition, and self-directed methods are surveyed for solitary study and homeschooling, with links to research and tools provided.
Log-concave distributions
The probabilist’s equivalent to convex optimisation objectives
Wherein the notion of log-concavity is introduced, and it is observed that the class is closed under affine maps, marginals and convolution, is unimodal, and is shown to admit a well-posed maximum-likelihood estimator.
Statistical learning theory for dependent data
Wherein non-stationary, non-asymptotic generalization bounds for time-series learning are presented, and sequential Rademacher complexity with a data-estimable discrepancy is employed to obviate mixing assumptions.
Efficient Langevin sampling Gaussian distributions
An educationally quixotic exercise
ML mavens who want a finger exercise in Langevin diffusions
Livescribe
A nifty smart pen that I use for stylus input despite various qualms
Wherein a ball‑point instrument is described, with its tip housing a camera that records strokes upon special Anoto‑patterned paper, with large onboard memory reducing sync frequency, and Bluetooth transfer to an app
Ensemble Kalman methods for training neural networks
Data assimilation for network weights
Wherein neural-network training is approached via ensemble Kalman updates, a dynamical-perspective method is presented, and a connection to stochastic gradient descent is examined through Claudia Schilling’s filter.
Terminals
Wherein various terminal emulators are surveyed and their traits are catalogued, and support for inline graphics via Sixels and tools like Chafa is noted, with platform and implementation differences delineated.
GP inducing variables
Wherein sparse Gaussian-process posteriors are approximated by inducing variables, Jensen’s inequality is applied to obtain a variational lower bound computable in O(m^3), and a trace term penalises residual covariance.
Verification and detection of generative AI
Watermarks, cryptographic verification of the products of AI
Wherein cryptographic signatures and subtle token-based watermarks for AI outputs are examined, including hash-based token selection that minimises Kullback–Leibler divergence, and robustness to adversarial removal is assessed.
Certification of neural nets
Watermarks, cryptographic verification and other certificates of authenticity for our computation
Wherein zero-knowledge proofs and proof-of-learning techniques for neural networks are examined, with iterative zkPoT constructions, robustness verification, and cryptographic separation of verification methods being considered.
LaTeX
…and ΤeΧ, and ConTeXt and XeTeX and TeXleMeElmo
The least worst mathematical typesetting system for the last 30 years and still for now. One of the better scoured of the rusting pipes comprising academic plumbing. De facto…
Subculture dynamics
Coalitions, scenes, fandoms, subcultures, normies, hipsters, sects, and tribes at scale
Wherein the formation and decline of peripheral groups is delineated, and recruitment, gatekeeping, and predation on MOPs is described as being amplified by the internet.
Bureaucracy
The theory of Iron Laws and Moral Mazes
Wherein the slow calculus of offices is surveyed, and the deferred repair of a coking battery is shown to cause far greater collapse, legal penalty, and expense than an earlier decisive investment would have averted.
Institutions for angels
Adverse selection in social movements
Wherein open‑membership movements are examined and shown to be vulnerable to elite capture and rage‑based online recruiting and clicktivism, with the shrillest voices becoming de facto spokespeople.
Fish shell
A command line shell that does not think that the problem is you
Wherein the command-line shell is presented with an opinionated design, its configuration is kept in ~/.config/fish/config.fish, and PATH manipulation is handled via fish_add_path and fish_user_paths.
Audiobooks
Wherein the landscape of audiobook suppliers and conversion tools is surveyed, and Audible‑centric workflows and practical scripts for concatenating chapterized MP3s into single M4B files are described.
Bayesian sparsity
Wherein various Bayesian sparsity techniques are examined, including Laplace, spike-and-slab, and horseshoe priors, and their implications for posterior sparsity and model selection are described.
Bayesian model selection
Wherein Bayesian model selection is examined, and the tension between posterior model averaging and pruning by marginal likelihood or cross‑validation for sparse regressors is described.
Editing images using code
Painting with a hammer, blindfolded
Wherein the command line is examined as a means to script image edits, and libvips is shown to stream and pipeline operations from disk, enabling AVIF thumbnail generation, while ImageMagick, GraphicsMagick and G'MIC are catalogued and animated GIFs are noted as fiddly.
Neural learning for geoscience
Wherein planetary-scale neural methods are described, physics constraints are emphasised, spherical data handling is treated, and existing geospatial tooling such as torchgeo is recommended for Earth-oriented spatiotemporal prediction at massive scale.
Data sets
Questions for answers looking for questions
Wherein a wide variety of public data sets are surveyed, and curated benchmark suites such as the Penn Machine Learning Benchmarks, SDMX‑accessible official statistics, and large social media archives are noted.
Capitalism’s end game
Wherein is examined whether human labour is displaced by automation and corporate consolidation, and whether resource depletion and low‑intensity conflict are induced as capitalism approaches systemic rupture.
Oral health
Wherein the engineered supertooth bacterium Lumina, xylitol’s purported anti‑caries effect, and a suggested association between bleeding gums and Alzheimer’s pathology in recent studies are presented for consideration.
The (Fisher) score function
Wherein the gradient of the log‑likelihood is presented and its variance is identified with the Fisher information, while applications to Langevin and Hamiltonian sampling are outlined, and conditional scores under nonlinear transforms are examined.
Langevin dynamcs MCMC
Wherein the Langevin SDE is invoked for sampling, its Euler–Maruyama discretisation with step ε and Gaussian innovation is described, and Metropolis adjustment and score-based perspectives are noted.
Safetyism and alignment problems
Wherein the prevalence of procedural safety measures, illustrated by requirements for event insurance and risk forms, is observed and the role of inaction in systemic harm is examined.
Foundation models for partial differential equations
Wherein foundation models for PDEs are examined, transformer-like architectures and token-based representations are queried, and the challenge of conditioning for inverse problems is outlined.
Diagrams
Of the kind I need, a practical guide to the creation thereof
Wherein tools and workflows for scientific line drawings are surveyed, with emphasis on producing vector‑exportable figures (SVG/PDF) and on the practical, browser‑based use of diagrams.net.
Neural learning for extreme values
Wherein neural methods for modelling extremes are examined, and the complications of spatial dependence and copula-linked correlations are set forth, while singular rare occurrences are noted.
Neural learning for spatiotemporal systems
Wherein neural learning for image-to-image regression of 2D fields is treated; stochastic PDEs are learned via U-Net and operator-learning methods, inverse problems are approached with low-rank or lattice Gaussian processes, and geospatial scales with out-of-sample spatial inference are examined.
Thailand
Wherein a year spent in Thailand in 2005–2006 is recorded, notes from a now‑lost blog are noted, and the country’s Buddhist meditation revival and uncolonized colonial history are outlined.
Graph neural nets
Neural networks applied to graph data. Neural networks, of course, can already be represented as directed graphs or applied to phenomena that arise from a causal graph, but…
Gender identity
Political and/or empirical engagement therewith
Wherein the social sorting of people into gender categories is examined, and the practical frictions in sport, medicalisation, and cross‑cultural negotiation are outlined, while correlations with autism and hormone debates are noted.
Monastic traditions
Wherein the social roles of monks are examined, with firsthand observation in Thailand noted, and pseudo-monastic practices, civic functions, and temple-based education are recorded.
Catastrophic risk
All your base are belong to dust
Wherein catastrophic risks are characterised as events that cannot in principle be indemnified by insurance, and it is noted that insurers and markets can be extinguished in a systemic collapse, with AI-driven failures and pandemics being adduced as examples.
Podcasts
Wherein podcasts are surveyed as aural companions, and a catalogue of shows—notably machine‑learning and history programmes with annotated killer‑to‑filler ratios, committed runs and hiatuses—is presented.
Reality gap
The difference between the real world and the simulations we use to model it
Wherein the distance between simulator and world is treated as an object of study, models are trained to characterise both simulator idiosyncrasies and physical measurements, and metrics are proposed.
Variational message passing with high-dimensional and functional nodes
Wherein variational message passing is extended to high-dimensional and functional nodes, and particle-filter-style ensemble methods are employed to represent and propagate approximate posterior messages over functions.
Classical model complexity penalties and dimension estimation
Information criteria, degrees of freedom etc
Wherein classical complexity penalties are surveyed and the relationships between AIC, Cp, SURE and BIC are expounded, and complications from non‑differentiable penalties such as LASSO are noted.
Inner experience in humans
Wherein the phenomenon of an inner monologue, its absence in some persons, the inability to form mental images (aphantasia), and the difficulty naming emotions (alexithymia) are described.
The public sphere and its business models
Free speech versus market-clearing speech
Wherein the public sphere is cast as a managed commons, attention is treated as a scarce resource, and Ostromian institutional remedies for discourse and advertising-funded debate are examined, including credentialed community norms.
Physical infrastructure
Wherein the hidden drivers of cost and emerging materials such as self‑healing concrete and mass timber are examined, and construction practices and infrastructural complexity are briefly catalogued.
Fonts
Typefaces I need, which is to say, for web pages and scientific papers
Wherein the complexities of self-hosting web fonts and selecting monospace designs for code are examined; methods for embedding Google fonts are outlined and a practical coding-font shortlist is provided.
Transferring money
Wherein international payment methods are surveyed and the limitation that Revolut’s crypto trades are non‑withdrawable in Australia is noted, and alternatives to bank transfer fees are outlined.
Quarto website preview server hacks
Wherein the quarto preview server is described as consuming gigabytes of RAM and is supplanted by a lightweight caddy file server run on 127.0.0.1:9889 for local testing
Posterior Gaussian process samples by updating prior samples
\[\renewcommand{\var}{\operatorname{Var}} \renewcommand{\cov}{\operatorname{Cov}} \renewcommand{\corr}{\operatorname{Corr}} \renewcommand{\dd}{\mathrm{d}} \renewcommand{\vv}[…
Confidentiality
Wherein a triage for personal privacy is presented, steps are recommended to keep bank logins from criminals and friends’ histories from corporate or state gaze, and beginner privacy tools are indicated.
Superstimuli
Somewhere around the Goodhart-Moloch-supernormal-alignment-utility-addiction area, we find the concept of superstimuli, things that feel better (or worse) for us than they…
Addiction
Wherein the brain’s opioid pathways and treatments such as naltrexone and semaglutide are examined, and the dampening of cravings for alcohol, shopping, or scrolling is described.
Overparameterization in large models
Improper learning, benign overfitting, double descent
Wherein the paradox of surplus weights is explicated and the double‑descent phenomenon in training error is examined, and the existence of compact subnetworks within massive models is noted.
Materials informatics
Machine learning in condensed matter physics, chemistry and materials science
Wherein a catalog of datasets, libraries, and projects for machine-learning–driven materials discovery is presented, including pointers to large computed databases such as the Open Catalyst datasets and the Materials Project.
Nearly-low-rank Hermitian matrices
a.k.a. perturbations of the identity, low-rank-plus-diagonal matrices
People with undergrad linear algebra
Ablation studies, knockout studies, lesion studies
In order to understand it we must be able to break it
Wherein the limits of inference by destruction are examined and it is noted that causal structure is not uniquely recoverable from final states, so training history is required for identification.
Fan fiction
Wherein the hothouse of transformative works is described, and AO3’s fifteen-year rise from a blog post to a major archive is recounted, with erotic and queer practices noted.
Artificial intimacy
Wherein the emergence of companion machines is described, and an empirical study is cited showing small cohorts are found to exchange outsized token volumes with chatbots, indicating emotional-dependence relations.
Trousers
Wherein the particulars of knee‑height fashions are examined, and a predilection for tweed and knickerbocker lengths, with tweed noted as scarce in Australia, is observed.
Fandoms
Wherein the vast imaginary realms of fanfiction are treated as theatres of proxy war, and the harassment of Kelly Marie Tran is recorded as an instance of online fandom violence.
Reproducing kernels satisfying physical equations
Wherein kernels are contrived so that reproducing-kernel Hilbert spaces yield solutions of prescribed PDEs, and operator-valued, divergence-free or latent-force kernels on domains, even the sphere, are exhibited.
Who I donate to
Wherein a catalogue of regular donations is presented, a 3% income giving level is disclosed, and a preference for high-risk, system-changing organisations, lobbyists, and open-source infrastructure is shown.
Medicalisation
Wherein the political economy of designating states as disease is examined, and the status effects, incentives, and iatrogenic harms of medical classification, as seen in ADHD and transgender identities, are delineated.
Implicit variational inference
Variational inference without densities
Wherein implicit variational inference is presented, models whose likelihoods are intractable are employed and KL-based losses are recovered via adversarial-style density-ratio estimation.
Sequential Monte Carlo
Wherein a population of samples is updated in nested stages to incorporate successive information, the method is presented as a generalisation of particle filters and is framed by interacting particle systems and Feynman‑Kac formulae.
Superintelligence
Wherein the prospect of machine minds is considered in sober detail, alignment and timelines are surveyed, and the practical question of whether humans would be pensioned by post‑human systems is raised.
Annealing in inference
Tempering, cooling, Platt scaling…
Wherein the temperature parameter is shown to scale log-densities, yielding tempered and cold posteriors and a data-weighting interpretation, as exemplified by covariance scaling in Gaussian models.
Krylov subspace iteration
Power method, pagerank, Lanczos decomposition, …
Wherein Krylov subspace iteration is presented as a method by which matrix inverses are approximated via a Lanczos Q T^{-1} Qᵗ factorisation, constructed from matrix–vector products at cost O(NDk + Dk³).
The occupation kernel method
Wherein occupation kernels for trajectories are presented, and their link to the Radon transform is noted, enabling identification of forcing fields from sparse observations without finite differences.
Function space versus weight space in Neural Nets
Wherein the contrast between function-space Gaussian process perspectives and weight-space optimization is examined, and the role of overparameterization and kernel limits is delineated.
Models for count data
Wherein count distributions are surveyed and unbounded support is assumed, Poisson and Negative Binomial families are exhibited, and mean–variance (dispersion) parameterizations and numerical issues are noted.
Cascade models
a.k.a. cluster distributions, Galton-Watson models
Wherein cascade population totals are considered for branching processes, generating functions are derived for geometric offspring, and total-size laws such as the Borel–Tanner and generalized Poisson are exhibited.
Kolmogorov-Arnold neural networks
Don’t learn weights, learn activations encode a physical process.
Wherein a neural architecture is described that learns univariate activation splines, is placed between symbolic regression and MLPs, and an N^{-4} scaling of test loss is reported.
High dimensional statistics
Wherein is described how probability mass in high dimensions is confined to thin annuli, low-dimensional projections are shown to rarely interpolate, and convex-hull interpolation thresholds are shown to scale like 2^{d/2}/d.
Gradient flows, sometimes stochastic
Infinitesimal optimization, SDEs and generalisation
ODE models (i.e. continuous limits) of optimisation trajectories, especially stochastic gradient descent. There are various flavours here, but I do not attempt to put them…
Divisible, decomposable and stable distributions
Ways of slicing randomness into easy chunks
Wherein the relations of divisibility, decomposability and stability are delineated and it is noted that infinitely divisible laws induce Lévy processes and admit Generalized Gamma Convolutions as representations
Rationalists in Australia and surrounding regions
Wherein networks of rationalists in Australia and neighbouring lands are catalogued, and their tendency to be organised chiefly via Facebook groups and meetups is noted, with local EA and AI‑safety chapters listed.
Bikes
Especially bikes where I live, which means Melbourne at the moment
Wherein the merits of commuting by bicycle are calculated and a personal saving of about AUD9,500 per year is reported, while equipment, routes, lights and theft precautions are surveyed.
Adaptive stochastic gradient descent
I need to mention Adam and RMSProp etc somewhere
Wherein the Adam variant of adaptive stochastic gradient descent is treated as a Bayesian update, and the peculiar square‑root scaling of gradient second moments is examined as a rationale for per‑parameter learning rates.
Moloch, slack and hyperselection
Weak selection versus goodharting, 内卷, involution
Wherein the inevitability of competitive selection is considered, the tension between ascendancy and reserve capacity is delineated, and platform enshittification is exemplified by TikTok’s progressive capture of value.
Symbolic regression
Wherein the recovery of a closed-form generating model from noisy, low-dimensional data is considered, and the task of learning model structure is cast as sparse, interpretable regression for discovery rather than parameter estimation.
Free content
Open source and public domain cultural resources
Wherein repositories of freely remixable media are herein catalogued, and Open Culture is cited with inventories of hundreds of free films, audiobooks, and ebooks for collage and illustration.
DIY VPN access point
I would like to use an anonymising VPN in my house. I could install separate software on each device, but this is unsatisfactory. By default, our household devices should…
Voting systems
Wherein it is declared that voting systems are surveyed and impossibility theorems, strategic manipulation, and alternatives such as preferential, approval, quadratic voting and sortition are examined.
Fisheries management
Wherein fisheries management is recorded as having been studied and the transferable‑quota system, in which a government‑set total allowable catch is apportioned into tradable individual shares, is noted.
Economic mechanism design
Designing markets and games to achieve what we collectively want from what we individually want
Wherein incentive mechanisms are treated as formal models into which numbers are plugged and computational complexity is derived, and transferable fishing quotas are examined as a concrete case study.
Ecological stability
The explosive growth and collapse of May’s paradox
Wherein May’s 1972 analysis is recounted, in which increasing species number and interaction density, when assigned randomly, are found to reduce the probability of systemic stability, and a paradox is posed.
Free books
Ideas in the age of mass reproduction
Wherein sources for freely accessible books online are catalogued, including library digitizations, textbook repositories, and ultra‑high‑resolution scans, and the Internet Archive’s lending program is reported as halted by major publishers.
Internet in Australia?
Wherein the national NBN rollout is treated as a political battleground and a recent switch to Aussie Broadband is reported, with customer service contrasted to prior ISP failings.
Decision theory
Wherein the calculus of choice is treated as consequence-driven, and it is noted that identical probabilities such as a 1% risk are weighed very differently by a farmer and a policymaker, and testing practice is reframed toward utility.
Epistemic bottlenecks
Wherein the transmissibility of knowledge is examined, transmission costs are framed through a Kolmogorov‑style minimal description, and the capacity of LLMs to teach other models is considered.
Public goods
Rival and excludable goods, you need to know what these are
Wherein the classification of goods by rivalry and excludability is laid out, and the particular challenge of funding non-excludable, non‑rival public goods such as national defence is examined.
VS Code for Python
An acceptable python IDE for very low cost
Wherein an interactive cell workflow using # %% is demonstrated, remote debugging via debugpy attach is explained, and fast linting with ruff is reported to be adopted.
Reproducible research
Wherein reproducible research is treated as the publication of code, data, and build pipelines, and experiments are encapsulated in containers and tracked with logs so that results are re‑run and re‑examined.
Let’s solve social event organising
Wherein the problem of arranging casual gatherings is considered, and calendar-link generation and niche invitation services are offered as means to coordinate dates and collect RSVPs.
Community
Engineering, maintaining, organizing, engineering oxytocin and dopamine…
Wherein methods for designing replicating movements are surveyed, and onboarding rituals are described, notably early editorial feedback that yields a shared Google Doc of comments as a first deliverable.
Free will
Let’s conflate stochastic dynamics with stochastic control
Wherein the question of human volition is debated over dinner, and emphasis is placed on predictability rather than metaphysical determinism as it is made relevant to legal responsibility, with a slot‑machine analogy adduced.
Python cluster computing
Parallel computing, wherein a head process spawns workers executing some python function
Wherein various Python cluster options are surveyed, and occasional SSH tunnel management is observed to be required, with joblib, dask, pathos and dispy being described as available approaches.
VS Code as LaTeX editor
Wherein VS Code is treated as a LaTeX editor, and its Overleaf integration and configurable build recipes such as latexmk and tectonic are detailed, with platform-specific SyncTeX setup discussed.
Project management
With special attention to speculative and innovative projects
Wherein the conduct of projects is treated as a study in institutional affordances and teamwork, and the peculiarities of innovative research projects and the planning fallacy are delineated.
ssh
Wherein the art of secure remote access is described, and the utility of SSH as a versatile tunnelling and identity‑handling tool, including ssh-agent, VPN‑like sshuttle, and config hardening, is delineated.
Spherical coordinates
Wherein the angular dependence of functions on the globe is resolved by the spherical harmonic transform, using degree ℓ and order m to decompose Laplace‑equation solutions.
Predictive coding, Free Energy in the sense of Friston
Wherein the free‑energy account of predictive coding is laid out, its variational Bayesian form is sketched, and the dark‑room paradox plus computable message‑passing implementations are noted.
Artificial life
Wherein the universality of life-like behaviour is examined through simulations of evolutionary algorithms, cellular automata, and related models, and the question of whether many different systems evolve replication and computation is posed.
Symbolic system identification
Wherein the recovery of governing equations from observational data is considered, and weak-form approaches and sparse Bayesian model selection, including SINDy and the forward-solver-free WENDy method, are described.
Modern conspiracy theorising
Lay theories of political science. This one is very much of the moment and is interesting for a variety of reasons, e.g. as weaponized social media strategy. Or, if you…
Rituals and beliefs that bind
Wherein belief systems are examined as social binders, and the role of coordination on falsehoods via Schelling points and low‑commitment online rituals is elucidated.
Audio synthesis in Python
Sometimes it is the right time to use the wrong tool for the job
Wherein Python’s methods for audio synthesis are surveyed and compared, including offline tools like paulstretch and real‑time frameworks such as pyo, and the practical expedient of invoking ffmpeg is reported.
The softmax function
Wherein real-valued vectors are mapped to simplex weights by exponentiation and normalization, the entropy of the resulting categorical distribution is derived, and its gradient is shown to be the probability vector minus one.
Health and chemistry
Wherein the persistence of PFAS in water, lead contamination of urban soils, and aircraft exhaust effects on air quality are documented, and maps and monitoring initiatives are cited.
Classification
Wherein the task of attributing observations to distributions A or B is treated, and a catalogue of classifier target losses is presented, including Matthews correlation coefficient, ROC/AUC, cross-entropy and expected cost.
Pytorch
The best-supported neural network framework
Wherein PyTorch is presented as the heir to Lua Torch, its eager dynamic graph model and clear documentation are noted, and an extensive ecosystem including functorch and Lightning is described.
Mathematica
A computational symbolic algebra system.
Digital scientific workbooks
The exploratory-algorithm-person’s IDE-equivalent. Literate coding meets science. a.k.a. dynamic report generation, a.k.a. literate programming.
Coarse graining
Wherein the systematic loss of detail under ordered information discarding is examined, renormalization and multi‑fidelity modelling are invoked, and persistent homology is noted as an analytical tool
Probabilistic numerics
Wherein numerical computation is treated as statistical inference and uncertainty from the computation itself is quantified, and priors over functions are employed to enable methods like Bayesian quadrature.
Stochastic quantization
Wherein stochastic quantization is examined as a diffusion-like procedure in field theories, and the denoising-diffusion trick is paralleled while an analytic target density is adopted in place of data-dependent priors.
pandoc
An itemised list of the esoteric difficulties involved in bullet points
Wherein pandoc is presented as a universal converter, written in Haskell, and is shown to convert markdown, LaTeX, HTML and MS Office formats while exposing filters and a JSON AST.
Meditation and enlightenment
Wherein meditation is treated as a method of mental training and empirical inquiry; EEG headbands and Quantified Self self-experiments are described as means by which effects are measured.
Functional reactive programming
Especially for user interfaces
Wherein functional reactive programming is described as the union of stream processing and near-functional techniques, and its use in user-interface design with ReactiveX libraries across languages is noted.
Editing images with machine learning
Wherein manual toil in image editing is relinquished to machine learning, and background removal, object erasure, and AI super‑resolution tools are presented as practical replacements and are offered by inexpensive web apps.
Righting and wronging
Harnessing righteousness for rightness
Wherein the social training of moral opprobrium is described, and taboos are portrayed as scarce instruments that are deployed to steer future behavior and to inhibit public deliberation.
Model mixing, model averaging for regression
Switching regression, mixture of experts
Wherein mixture priors are considered and the posterior is exhibited as a weighted mixture of component posteriors, the weights being updated by each component’s marginal likelihood.
Home networks for lazy dummies
Wherein practical home-network choices are outlined, including guest VLAN setup and flashing a router for a VPN AP, Australian ISP options are referenced, and Ubiquiti update risks are noted.
Sampling by the method of the ratio of uniforms
Wherein the ratio-of-uniforms technique is described, and a planar two-dimensional region is employed to generate random variates by accepting points inside a constructed envelope.
Probabilistic programming
Doing statistics using the tools of computer science
Probabilistic programming languages (PPLs). A probabilistic programming system is a system for specifying stochastic generative models and reasoning about them. Or, as Fabian…
Perception and evolution
How much of reality is it worth it for us to see? Umwelt, etc. Biological phenomenology.
Wherein the likelihood that organisms' perceptual representations correspond to external structures is examined by appeal to evolutionary and decision‑theoretic considerations, and adaptive fitness consequences are considered.
Simulating climate
Wherein climate simulations are described, with machine learning applied to atmospheric, oceanic, and glacial processes, an instructed glacier model is cited, and open datasets and resources are catalogued.
Atmospheric modelling
Wherein atmospheric modelling is presented with marine-tornado and storm imagery, ML-enhanced methods are described for spatiotemporal analysis, and flux inversion is applied to trace pollution sources.
Vector databases
Wherein vector indices for similarity search are described, and HNSW and other nearest‑neighbor algorithms are noted as enabling billion‑scale retrieval and embedding‑based memory for AI agents.
Oceanography
Wherein the ocean is treated as a spatiotemporal system and the modelling of surfzone and tsunami waves by the FUNWAVE‑TVD fully nonlinear Boussinesq solver with shock‑capturing breaking and wetting‑drying is described.
Reconciliation of overlapping Gaussian processes
Combining Gaussian processes on the same domain; consistency; coherence; generalized phase retrieval
Wherein two Gaussian process priors defined on a common domain are reconciled by forming a product-of-densities posterior, and the spatially varying uncertainty and ties to Griffin-Lim iteration are examined.
Weighted data, weighted likelihoods in statistics
Wherein the distinctions among precision, frequency, and sampling weights are laid out, variance implications are contrasted, and iteratively reweighted least squares is noted as linked to robust estimation and Bayesian tempering.
Gamma distributions
Wherein the Gamma distribution is presented in the shape–rate parameterisation and is described as having mean α/λ and variance α/λ^2, with closure under addition occurring only for identical rate parameters.
Monte Carlo gradient estimation
Wherein Monte Carlo gradients are treated via score-function (REINFORCE) estimators and reparameterization through a base distribution, categorical cases are handled by Gumbel‑softmax, and inverse‑CDF differentiation is considered.
Conjugate priors
Wherein conjugate priors are presented for exponential‑family likelihoods, and posterior parameters are shown to be updated by adding sufficient statistic T(x) to λ and incrementing ν by one.
LaTeX mathematics hacks
Wherein a variety of LaTeX math hacks are presented, spacing and line-break quirks are surveyed, a \numberthis macro for selective equation numbering is explained, and independence-symbol hacks are demonstrated.
Temporal generative adversarial networks
Wherein temporal generative adversarial networks are employed as time-series predictors and are cast as stochastic differential equations for forecasting, while connections to path-signature methods are noted.
Functional stochastic differential equations
SDEs taking values in some function space
Wherein the extension to infinite-dimensional Wiener processes is treated, and the distinction between cylindrical and Q‑Wiener perturbations is explicated, with functional noise being incorporated into evolution laws.
Soft incentive mechanism design
Markets, cakes, karma, and reverse game theory, applied to humans
Wherein the strict axioms of mechanism design are relaxed to accommodate human moral wetware, and qualitative, fuzzy incentive models are proposed for institutions such as voting and reputation systems, and implications for governance are examined.
Inverse Gaussian distribution
Wherein the inverse Gaussian is presented as a tractable non‑negative exponential family with mean μ, a bivariate conjugate prior is given, and an explicit modified‑gamma marginal for λ is derived, with Lévy links noted.
Historical English
Wherein the author’s amateur forays into etymology, recreational word‑history, Anglish experiments, and Chaucerian pronunciation are surveyed, and 17th–19th‑century cant dictionaries and a podcast are noted.
Academic reading workflow
Wherein the common PDF is described as incompatible with e‑readers, and procedures for annotating, syncing annotations to citation managers, and employing AI tools for literature queries are surveyed.
Curved exponential families
Wherein families are characterized by a natural parameter vector that is constrained to a curve in the canonical parameter space, so the sufficient statistic dimension exceeds the parameter dimension and k>q is exhibited.
Music gear
Pragmatic digital amps and microphones
Wherein in‑canal Shure SE215 earphones are recommended for travel, a QSC K.2 rig is noted to run from a 10A circuit, and RME Fireface class‑compliant modes are explained.
History of medicine
Quacks, Mountebanks, Big Pharma
Wherein the protracted misapprehension of airborne transmission is examined, and a sixty-year methodological error that aggravated the COVID pandemic is set forth as a concrete instance.
DIY home repair stuff
Wherein practical home repairs are surveyed and the presence of asbestos is noted, the fact that lead paint was never fully banned in Australia is recorded, and methods for hanging fabric wallpaper are outlined.
Video editing
Wherein inexpensive and free/libre video editors are surveyed, a web-based editor Pikimov is noted, transcript-driven editing via Descript is described, and live compositing with OBS is offered.
Exponential families
Wherein the structure of canonical parameters and sufficient statistics is laid out, conjugate priors are exhibited, log‑partition functions are related to moments, and tempered powers are considered for normalization.
R
The statistical programming language, not the letter
tl;dr R is a powerful, effective, diverse, well-supported, free, messy, inefficient, de-facto standard. As far as scientific computation goes, this is outstandingly good.
Bayesian model selection
Wherein models of varying dimensionality are traversed via reversible-jump Markov chain Monte Carlo, acceptance probabilities being computed with Jacobian adjustments for dimension-changing proposals.
Bandit problems
Wherein the formal problem of sequential choice is presented, and the exploration–exploitation trade-off in learning optimal policies from limited rewards is examined, including contextual and adversarial flavours.
Non-stationary bandit problems
Wherein the case of bandits with time-varying rewards and restless arms is set forth, foundational results of Whittle are invoked, and change-detection mechanisms and upper-confidence procedures are considered.
How to reduce corporate spying
Wherein the ubiquity of corporate surveillance is described, including how modern macOS is shown to transmit hashes of run applications unencrypted to third‑party CDNs, rendering activity observable to network actors.
Transformer networks for generic time series prediction
Wherein transformer architectures are applied to multivariate, irregularly sampled sequences, attention mechanisms are adapted to handle missing timestamps, and model performance is measured on benchmark datasets.
Kernel distribution embedding
Conditional mean embeddings etc
Wherein kernel embeddings of probability measures are presented, and it is shown how conditional embedding operators, estimated from samples via Gram matrices, are used to express priors and posteriors by a kernel Bayes’ rule
Niche construction
Wherein organisms are described as altering their surroundings, earthworms are shown to engineer soil that favors their kind, and selection is recast as acting upon phenotypes rather than solely genotypes.
Tip me
Wherein modest support is solicited by means of PayPal, Brave browser tips, cryptocurrency musings, and purchases of music or printed wares, and the author’s acceptance is recorded without ceremony.
Normalising flows
Wherein normalising flows are described as compositions of invertible, differentiable maps from a simple base (often Gaussian), and densities are obtained by Jacobian‑determinant calculations to enable variational inference.
Neural mixtures of experts
Switching regression, mixture of experts
Wherein a consortium of expert networks is combined by a learned gating function that is trained to allocate inputs to specialists for distinct input regimes, and model selection is performed during inference.
Navigating large organisations
Wherein methods for taming bureaucracy in large organisations are set forth, including keeping work parallel and employing 360° reviews and off‑the‑shelf feedback tools to surface coordination friction.
Community governance
Coordinating human beings by collective application of social force
Wherein community governance is surveyed and models such as sociocracy and transformative justice are examined, and federated tools like Mastodon and Loomio are offered as coordination mechanisms.
The Gaussian distribution
The default probability distribution
Many facts about the useful, boring, ubiquitous Gaussian. Djalil Chafaï lists Three reasons for Gaussians, emphasising more abstract, not-necessarily generative reasons.
Uncertainty quantification
Wherein machine learning predictions are furnished with quantified confidence, and the distinction between aleatoric and epistemic uncertainty is explained, with conformal prediction mentioned.
Teamwork
Wherein psychological safety, small parallel work units, and localised resources are presented as determinants of team effectiveness, and methods for fostering group flow and communication are recorded.
Learning with PDE conservation laws
Wherein neural networks are engineered to enforce conservation symmetries via architectural constraints rather than by loss penalties, thereby preserving PDE invariants and being contrasted with PINN approaches.
Genre speciation
Wherein genre speciation is treated as a phylogenetic process, and music and religion are traced through data-driven maps and sectal trees, including Spotify’s Every Noise at Once as a concrete example.
Noise outsourcing
Wherein a conditional law is represented as Y = f(η, S(X)) with η taken independent and uniform, and a separating statistic S is used to outsource randomness for reparameterization and symmetry-aware learning
Deep sets
Invariant and equivariant functions, learning to aggregate
Wherein exchangeable functions are examined, it is noted that a permutation‑invariant set function is represented as ρ of the sum of φ(x), and connections to self‑attention and upper bounds on set size are recorded.
Judaism, Jews in history, art and culture
Wherein the visual record of Jewish life is surveyed through historical prints catalogued by the Rijksmuseum; a depiction of Moses tapping the rock is appended as an illustrative plate, and prints from the seventeenth to nineteenth centuries are represented.
(Kernelized) Stein variational gradient descent
KSD, SVGD, other computational Stein discrepancy methods
Stein’s method meets variational inference via kernels and probability measures. The result is a method of inference that maintains an ensemble of particles which notionally…
Functional programming
Wherein computation is treated as the evaluation of mathematical functions, mutable state is avoided, functions are allowed as values, and their use in differentiable and probabilistic languages is noted while memory reuse is studied.
Messenger shooting
Distinguishing bad things from bad news
Wherein the mechanics of blame and messenger-targeted retaliation are examined through the lens of predictive coding and attribution in reinforcement learning, with accompanying religiously inflected illustrations.
Death
Wherein the ritual of companioning at life’s end is described, the medieval Dance of Death’s illustrated tradition is noted, and the modern role of death doulas in bedside care is mentioned.
Nested sampling
Wherein the estimation of Bayesian evidence by a Monte Carlo method is described in dispassionate terms, and contemporary implementations in JAX and probabilistic programming are catalogued.
Doubly robust learning for causal inference
TMLE, debiassed ML, X-learners, Neyman learning, Targeted learning
Wherein a framework is described in which machine‑learning base learners—random forests, BART, neural networks—are used to estimate the conditional average treatment effect via double‑robust orthogonalization.
Agtech/Agritech
Dotcomese for “agricultural technology”
Wherein microclimate sensing and machine intelligence are employed in Australian enterprises, and livestock management software is catalogued as part of technologized food production.
Basa Sunda
Wherein the Sundanese tongue is presented, and attention is directed to its revived script now encoded in Unicode, local honorifics such as Mang and Kang are noted, and a Kawali stone inscription is cited.
Bayesian model selection by model evidence maximisation
Type II maximum likelihood, marginal maximum likelihood, Bayes Occam’s razor, Bayes factor
Wherein model selection is conducted by maximising marginal evidence, the Bayes factor is invoked as the decision criterion, and marginal evidence is shown to be computed from integrated likelihoods.
Jensen Gap
Wherein the difference f(E[X])−E[f(X)] is considered as the Jensen gap, and bounds for its magnitude are presented for differentiable continuous functions and for discrete distributions.
Entropy vs information
MaxEnt(?), macrostates, subjective updating, epistemic randomness, Szilard engines, Gibbs paradox…
Wherein the relation between thermodynamic entropy and informational measures is examined, MaxEnt priors are considered, macrostates are framed as Markov partitions, and algorithmic complexity links are noted.
Big data ML best practice
Being transparent about what I put in this black box
Wherein guidance is set forth, emphasizing that despite vast datasets, experimental models remain small-data due to compute cost, and tools and testing workflows are described.
Transforms of Gaussian noise
Delta method, error propagation, unscented transform, Taylor expansion…
Wherein a nonlinear mapping of a Gaussian process is examined, and Taylor expansions and sigma‑point (unscented) moment approximations with explicit second‑order corrections are presented for propagated mean and covariance
Expectation maximisation
Wherein an optimisation method is presented, in which expectation and maximisation steps are alternated under a completed‑data model, and missing data are imputed via latent‑variable conditional expectations.
Pornography and other lewd art
Morsels of oddity from the depiction of human sexual behaviour
Discussion of sexual acts, depictions of sexual acts
Reproducibility in Machine Learning research
Wherein methods for verifying machine learning claims are examined, with emphasis on checklists and reporting standards such as the REFORMS 32‑item guide, and challenges of secretive foundation models are considered
Squeaky wheel equilibria
Mismatches between vocal-ness and collective desire, NIMBYism, intolerance
Wherein the preeminence of vocal minorities is examined, and the role of persistent public noise—manifested in street commerce and protest—is shown to determine prevailing social standards.
Editors for LaTeX
Wherein a menagerie of LaTeX editors is catalogued, and the ubiquity of SyncTeX alignment and WebAssembly in‑browser compilation is noted as practical distinctions among the rival tools.
Pyro
Approximate maximum in the density of probabilistic programming effort
Wherein a probabilistic programming language is described, implemented atop PyTorch or JAX, and shown to support modern inference such as NUTS and automatic graphical‑model rendering.
Collectivism and individualism
Wherein collectivism is presented as an axis of cultural difference, and a link to rice cultivation and household co‑residence of ageing parents is described, while interpersonal suspicion is reported.
Online shopping
Wherein strategies for reshipping are presented and browser‑isolated coupon and cashback apps that trade user data for discounts are examined, with testing noted and privacy trade‑offs highlighted.
Rough path theory and signature methods
Wherein path signatures, expressed as collections of iterated integrals of a path, are presented and are applied as universal feature transforms for sequential data, with ties to Gaussian-driven rough differential equations.
Physically-grounded computing
Wherein the limits of computation are surveyed, and the thermodynamic cost of information, via fluctuation theorems and entropy production in physical devices, is examined.
Internet search engines
Wherein the decline of mainstream search results is noted, and privacy‑centred paid services, DIY proxies, and AI‑augmented or decentralized search approaches are surveyed and contrasted.
Attention Deficit (Hyperactivity) Disorder
Wherein the pharmacology of stimulant treatments, the attention‑economy niche of the condition, and experimental findings of slower acquisition combined with faster extinction in learning are set forth.
The property market is pollution
Or: hunting endangered big game in dwindling habitat. Or: Investing in Boomer Bitcoin
Wherein the property market is portrayed as pollution, and years of household labour are shown to be siphoned into servicing mortgages, leaving civic time and communal life progressively eroded.
Algorithmic statistics
Probably also algorithmic information theory
Wherein the boundary betwixt determinism and chance is considered, and Kolmogorov complexity is invoked to treat deterministic processes as if stochastic, with attention paid to empirical detection of computation.
Guesstimation
Fermi calculations, calculation hacks, principled estimates, informed guesses
Wherein back-of-envelope Fermi methods are treated alongside probabilistic spreadsheets, Monte Carlo error estimation is employed, and an embeddable Squiggle language for JavaScript is described.
Git GUIs
Wherein various Git graphical clients are surveyed, their visualisations and heavy RAM footprints are compared, console and web-based alternatives are catalogued, and pedagogical usefulness is noted.
Optimisation
Wherein a brief survey of continuous and combinatorial methods is presented, and gradient‑based, gradient‑free, manifold and ADMM techniques are outlined for large‑scale and experiment‑bound problems.
An orderly retreat from economic relevance
Red-teaming short-term human purpose
Wherein human labour is considered in an age of automata, attention is drawn to embodied tasks and political roles that are harder to automate, and the cost of producing a working human is noted at about $100,000.
Resilience (Psychological)
Wherein the means by which mental well-being is cultivated are catalogued, and the problem of calibrating an optimum amount of challenge for lasting psychological resilience is examined.
Ethical consumption
Veganisms, boycotts and other individual solutions to structural problems
Wherein ethical consumption is examined and the limited cost‑effectiveness of consumer choices is argued, and systemic tools such as taxation and bans are described as alternatives to individual action.
Air pollution
For a new century in which inhaling gets only more dangerous
Wherein the effects of urban and indoor pollution are examined, including acute impacts on classroom cognition, household gas‑stove emissions, and mask filtration options for PM2.5 exposure.
Decentralized net services
a.k.a. web3, DEX, P2P, Peer-to-peer, friend-to-friend; Internet for an untrustworthy world
Wherein peer-to-peer internet services are surveyed, with examples from Bluetooth mesh messaging to IPFS file networks, and IPFS is observed to drain laptop batteries when run continuously.
Bayesian posterior inference via optimization
Wherein stochastic gradient descent is treated as a Markov chain and constant‑rate SGD, natural‑gradient Adam variants, and stochastic weight averaging are shown to be cast as posterior samplers under a quadratic‑mode approximation
Sufficiently good hedonism
Trade-offs between limited time, limited cash and limited imagination, toward the most happiness
The Good Life as far as I recognise it. Or: Training myself to recognise good lives.
Squared neural families
Wherein a class of neural models is squared and is presented as a generalisation of exponential families, while their statistical structure is recast in terms of classical parametrisations and sufficient statistics.
Contemporary epidemiology of mental health
Healthy norms, trauma, contagion, psychiatrisation, prevalence inflation hypothesis
Wherein the rise of self-reported disorders is examined through the lens of social media attention economies, and the hypothesis that extensive messaging partly inflates measured prevalence is considered.
Score matching
Wherein a method for learning the data score is described, its use in neural denoising diffusion models is exhibited, and a suggestive connection to thermodynamic interpretations of perturbed data dynamics is noted.
Designing less toxic social media
Wherein an iterative design process for less toxic social networks is proposed, with attention paid to recommendation algorithms, Indieweb and Fediverse approaches, and modest interface shifts to reduce harm
Bayes linear methods
Some kind of approximate Bayes thing
Wherein posterior updates are effected via means and covariance matrices without presuming Gaussianity, Matheron-style least-squares estimators are recovered, and belief adjustment is formalized.
Presentation tools
Slide decks, “powerpoints”, beamer lore
On tools for presentation.
Hardware for neural networks
Neuromorphic computing, non-von-Neumann architectures, and other ways to compute for AI
Wherein neural computation is surveyed through hardware modalities, and optical processing using randomized linear algebra and direct feedback alignment is noted as a concrete alternative to GPU backpropagation.
Variational message-passing algorithms in graphical models
Wherein variational message-passing is presented under KL divergence and exponential-family assumptions, on factor graphs assumed to be trees, and an application to latent Gaussian process models is sketched.
Rituals
Synthetic biology for utopian egregores
Wherein communal rites are examined, and a secular Lent food‑fight photograph is presented as an instance of ritualized social bonding outside doctrinal faith, while modern practices from Burning Man to psychedelic ceremonies are noted.
Unconferences
Hackfests, Open Space Technology, BarCamps, and other self-organising events for science purposes
Wherein the rise and eclipse of unconferences is recounted, an Open Space session in Stockholm with Prof Kelly Snook is described, and the dependence on skilled facilitators, plus academic metricisation, is considered.
Microstressors
Asymmetric interactions in aggregate, microinequity, microaggressions
Wherein the accumulation of frequent, mundane queries by coworkers about one’s background is counted, so that in a 100‑person firm a visible minority is shown to spend roughly eight hours explaining themselves.
Learning curves
Wherein a ten-year ascent for mastery of musical instruments is traced, individual and organisational paths are compared, and social impediments to skill diffusion are noted.
Matrix algebra
Maybe also some operator algebra
Wherein matrix algebra is treated as the non-commutative calculus of linear operators, infinitesimal matrices are admitted for matrix calculus, and symbolic tooling such as SymPy and Sage is surveyed.
Cognitive Enhancement
Wherein the practice of pharmacological brain enhancement is described as common among competitive technologists and financiers, modafinil use being noted alongside anecdotal side effects and legal uncertainty.
Stein’s method
His eyes are like angels but his heart is cold / No need to ask / He’s a Stein operator
Wherein Stein’s method is presented as a device for solving the Stein equation, and bounds are obtained via exchangeable pairs, with particular attention to Gaussian characterising operators and generator constructions.
Dynamic causality
Wherein the notion of treating streams as generated by small evolving graphical models for streaming recommender-systems is examined, and the causaloid concept of Hardy is invoked.
Cryptographic tokens, distributed ledgers, and blockchain-like-things
When all you have is a hasher…
Wherein distributed ledgers and cryptographic tokens are surveyed, and the massive energy consumption of proof‑of‑work consensus is noted as a salient practical consequence for their deployment.
Simulation-based inference
If I knew the right inputs to the simulator, could I get behaviour which matched my observations?
Wherein the problem of inferring simulator parameters without access to a likelihood is examined, and approaches are delineated, including auxiliary‑model matching, neural likelihood surrogates, and MMD‑based discrepancy measures.
Comment systems for static websites
Minimum viable interaction for the web
Wherein the practice of adding commentary to static sites is examined, and the option of using GitHub Discussions via giscus to store and fetch comments without a separate server is considered.
Let’s try substack
Wherein notebook entries are proposed to be cross-posted to an email letter via Substack when events are deemed momentous, and the tension with a continually revised notebook format is presented.
China
Wherein the centre of the world is surveyed and the Great Firewall’s evasions, United Front diaspora tactics, and podcasted historiography are recorded in a clipped, impersonal chapter-heading register.
Maximum Mean Discrepancy, Hilbert-Schmidt Independence Criterion
Wherein, by means of kernel embeddings into a Hilbert space, distributional differences are measured by a sample‑estimable integral probability metric, and dependence is tested via the Hilbert‑Schmidt criterion.
Maximum Mean Discrepancy flows
Wherein transport maps are constructed via Maximum Mean Discrepancy, and a denoising-diffusion-like iterative sampler is recast as an MMD-driven gradient flow linked to Wasserstein geometry.
Nostr
Wherein a simple relay-based protocol for decentralised social media is described, client implementations such as Damus and web relays are cited, and the absence of a blockchain requirement is asserted.
Factor graphs
Wherein factor graphs are presented as bipartite or Forney-style diagrams in which factors and variables are arranged so that local message‑passing is automated for marginal inference, and plates are unrolled into replicated nodes.
Institutional alignment problems
Mechanism design for distributed moral wetware
Wherein institutional alignment problems are examined through mechanism design for formalized distributed moral wetware, and the risk of excessive bureaucracy or mission failure from rule-bound agents is sketched.
Morphogenesis
Wherein differentiable cellular automata and dynamical-systems tools are applied to instructing cells toward differentiated bodies, and geometric organizing structures in gene-expression space are examined.
Intimate question systems
Conversation starters, truth-or-dare
Wherein intimate question sets are catalogued for relationship work, including the 36‑questions protocol, a life‑review variant, and commercial card games, and cards are observed to be colour‑coded to mark sensitive topics.
Microjoys
Gratitude lists, morale boosts
Wherein the neologism microjoys is introduced as a practice of noting small pleasures—bass-driven movement, crisp-cream textures, shared flow states—and is attributed to Cyndie Spiegel in the author’s milieu.
Battlers
Those who get the rough end of the pineapple, nanny states, paternalism, luxury beliefs, class wars
Wherein is examined how elite status displays and luxury beliefs are linked to drug reform and open‑borders advocacy, and how risk and social contagion are passed to wage‑dependent members of society.
Terminal session management and multiplexing
Wherein the practice of detaching processes to survive transient SSH disconnections is described, and contrasts between session‑only tools like abduco and multiplexers such as tmux are noted.
Social organisation of knowledge
Wherein the workings of news media and scientific circles are examined, with attention paid to misinformation, Bayesian design approaches, and impacts on public shared reality during elections.
Drugs, prescribed
Wherein the economics of antibiotic development are examined, and the tension between short treatment courses and long-term bacterial resistance is outlined, with market incentives shown to be misaligned.
The money laundry
Wherein the unraveling of illicit funds is described, and accountants are placed as instruments of investigation while simple payroll schemes and Nevis offshore registries are shown to enable concealment.
Typst
Wherein the typesetting system is presented as a LaTeX reimagining, is noted for built-in markup and incremental compilation, and is provided with a public web app that currently outputs PDFs only.
Status
Wherein the distinction between dominance and prestige is set forth, the role of prestige economies in generating cooperation beyond kin is observed, and reputation systems and phase transitions in hierarchies are noted.
Tensorboard
Wherein an experiment-tracking interface is described, its default visualisations being runnable on minimal servers and its filesystem-written logs being employed to circumvent HPC network lockdowns.
Tracking experiments in data science
Wherein the landscape of experiment tracking is surveyed; special attention is given to neural nets as ongoing convergence monitored, and tooling from Sacred to MLflow is catalogued.
Ethnomusicology
Wherein miscellaneous scraps on culture and musicality are assembled, and attention is directed to technoethnomusicology, Sundanese musical practices, and the Scotch Snap rhythm as a cross‑cultural datum.
Symbols and the public sphere
Sovereign citizens, LARPing insurgency, institution building, symbolic capitalism
Wherein the politics of symbols are examined as attention-drawing theatrics are contrasted with the swift creation of local emergency committees in 1918, and performative rebellion is traced to semiotic diversion.
Linear algebra
If the thing is twice as big, the transformed version of the thing is also twice as big. {End}
Wherein the foundations of linear algebra are surveyed, and the singular value decomposition is invoked so that linear maps are exhibited as data-approximation operators, with a brief note on Moore–Penrose pseudoinverses.
Mass power generation, engineering and use
On the magical, terrible requisites of keeping the lights on
Wherein early solar thermal generators are recorded in a 1916 Popular Science write-up and photovoltaics are traced to the 1880s, and a Distribution System State Estimation method is noted as commercialized by GridQube, while grids, nuclear costs and DIY solar are compiled.
Hyperparameter optimization
Replacing a hyperparameter problem with a hyperhyperparameter problem, which feels like progress
Wherein the practical contest between random search, Bayesian surrogate methods, and adaptive schemes such as Hyperband is surveyed, and toolchains for experiment tracking, early stopping, and HumpDay comparisons are noted.
Static websites
Websites that are just files on a server, which is all I need
Wherein the website is described as a folder of files on the hard drive, which are processed by a static site generator into HTML and are deployed to simple hosts, for example via Netlify’s drop service
Web browser hacks
Wherein browser containers are described as a means to isolate identities and sessions, single-site browsers are noted for dedicated app-like access, and userstyles are cited for interface control and privacy enhancements.
Structural programs are hard, let’s do training programs
“The least you can do” is the minimum unviable product
Wherein organisations are treated for systemic failings by the imposition of mandated training schemes, with implicit‑bias tests and wellness programmes being deployed while structural reform is declined.
Offline email syncing
Wherein local mailboxes are described as being served over IMAP or stored as Maildir/mbox files and are synchronized offline using tools such as isync or OfflineIMAP to permit local reading.
Python, compilation and acceleration of
Wherein various Python acceleration strategies are surveyed, and the generation of self-contained C++ by Pythran and JIT compilation to GPUs via Numba are described.
Gradient descent, Newton-like
Wherein second‑order curvature is examined and the role of Hessian approximations—Gauss‑Newton, generalized Gauss‑Newton and quasi‑Newton secant updates—are described as underpinning trust‑region and line‑search methods.
Indonesian music
Wherein Indonesia’s musical traditions are presented with gamelan’s tuned gongs traced into contemporary scenes, where recordings and distortion pedals are used to rework island sounds.
Hugo
Lightweight static site utility which is as fast and elegant as it is stubborn
Wherein the static site generator is presented as a compact, fast binary that avoids large node dependencies, yet is reported to require tedious workarounds for mathematics and to limit plugins via backends
Fashion
Theory and practice of hip garments
Wherein traditional tweed waistcoats and Indonesian batik provenance are examined, payment frictions with Indonesian vendors are noted, and upcycling, masks, and bolo‑tie artisanship are catalogued.
Javascript user interfaces
Wherein JavaScript user interfaces are surveyed and simple slider‑driven tooling such as dat.gui is recommended for rapid prototyping, and performance pitfalls like scheduling and jank are noted.
Data cleaning
Wherein data cleaning is presented as a laborious process, in which weak‑labeling methods like Snorkel and metrics such as Variance of Gradients are employed to surface and correct problematic examples.
Switching to netlify
Wherein the site is transferred to Netlify, its content is migrated from a blogdown backend to a Quarto backend, the build pipeline is adjusted, and intermittent disruptions are acknowledged as ongoing.
Metis and .*-rationality
High modernism, spontaneous order, legibility, the Great Society, technocracy, local knowledge
Wherein the persistence of seemingly irrational rites is explained by social coordination: a bullet‑proof powder is shown to lower perceived costs of resistance, thereby spreading by group selection.
Iterated and evolutionary game theory
Wherein the evolution of cooperation in populations is examined through the iterated Prisoner’s Dilemma, and the influence of population structure and memory, exemplified by Tit‑for‑Tat, is considered.
Recursive identification
Learning forward dynamics by looking at time series a bit at a time
Wherein recursive identification of dynamical systems is surveyed and a two‑step forward—one‑step back pushforward trick for mitigating distribution shift is presented, with attention to hidden states.
Economic inequality
Return to capital versus return to labour versus return to state of nature
Wherein the role of corruption in eroding institutional and social trust is traced, and the indirect links from entrenched elites and stalled growth to conflict and lost innovation are laid out.
Internet for the marginally online
Wherein remedies for intermittent connections are prescribed, mosh and robust wget patterns are recommended, and macOS wifi toggling and Finder workarounds are outlined for those with poor infrastructure.
All we need is hate
Wherein the utility of manufactured enemies is considered, and the grim calculus of selecting minimal scapegoats and employing dog‑whistled intolerance to buy in‑group cohesion is examined.
Single subject experiments
Instrumentation and analytics for body and soul. Quantified self. DIY precision medicine.
Wherein single-subject experimental methods are surveyed, and smartphone health exports, wearable biomarker streams, and methods for self-blinding are noted as specific avenues for measurement, logging, and analysis.
Stochastic Taylor expansion
Polynomial approximations of small randomnesses, Itô’s lemma
Wherein an Itô-based local expansion is presented with differential operators L0 and L1, and a note is made on higher-order iterations and extensions to Lévy noises, while practical use is deferred.
Narrative
Wherein the mechanisms by which narrative alters empathy and cognition are examined, evidence from neuroscience and media studies is surveyed, and its deployment in gamification to steer behavior is described.
Music software
A list of things that I have used or wish to try using to make sound come out of my computer. NB, this area is rapidly moving as AI moves into music; I have not updated it…
Software audio routers
Patch cables for the cable-averse
Wherein application audio is routed between isolated programs by virtual cables, and streamer-focused solutions are observed to have eased tedious setup, while macOS and Windows tools such as BlackHole and VB Audio are catalogued.
Audio sample libraries
Wherein the varieties of gratis musical samples are surveyed, and public‑domain field recordings, retro soundfonts, and convolution impulse‑response libraries are cited as concrete sources for reuse and synthesis.
Travel checklist
Wherein items for travel in Southeast Asia are enumerated, and measures for exercise and hygiene are provided; a collapsible water tank is carried for improvised weight training, and a local extension cord is included.
Hallucinations
Wherein the neurology and phenomenology of chemically or medically induced visions are described, and recurring mathematical motifs such as fractals and hyperbolic geometry are adduced.
Microsoft Windows for the avoidant
Wherein modern Windows is described as exhibiting built‑in advertising, and practical steps are provided for disabling ads, installing WSL and package managers, managing codecs, and a BitLocker failure is reported.
Audio/music corpora
Wherein datasets of raw audio and MIDI are catalogued, access to multitrack stems and tempo annotations is emphasized, label noise from crowdsourced sources is noted, and the Free Music Archive is recommended for whole songs.
Generative music with diffusion models
Wherein diffusion models are applied to musical audio generation, and text-conditioned controls are described, with tempo-aware conditioning and CLAP-based labeling being considered.
Effective altruism
Wherein the movement’s utilitarian calculus is examined, opportunity costs are foregrounded, and attention is given to low‑variance GiveWell recommendations, hits‑based giving, and long‑term AI‑risk priorities.
Moral wetware
What ethical operating systems can be executed on our neurosocial substrate?
Wherein humans are treated as moral wetware whose social brains and cooperative subsystems are examined for the learnability and generalization of ethical systems, with attention to permissive, nourishing institutions
Natural language processing software
Wherein contemporary toolkits for processing human speech and text are surveyed, and prominent libraries such as HuggingFace, spaCy and Gensim are catalogued for tokenization, embeddings and pipeline integration.
Computational Fluid Dynamics
Wherein the Navier‑Stokes equations are introduced as prototypical PDEs, computational approaches are catalogued, and differentiable solvers and graphics‑oriented engines for ML coupling are surveyed.
System identification in continuous time
Learning in continuous ODEs, SDEs and CDEs
Wherein continuous‑time system parameters are sought by methods adapted to stochastic differential equations, adjoint methods are examined, and modern solvers such as JAX/Diffrax are invoked, including sparse identification and PDE extensions
Python caches
The fastest code is the code you don’t run
Wherein disk-backed Python caches are surveyed, and a focus is placed on libraries that provide multiprocess-safe locking and memory-mapped storage permitting gigabyte-scale binary blobs.
Learning graphical models from time series
Also, causal discovery, structure discovery in time series
Wherein causal links among multivariate time series are sought by statistical discovery, attention being given to linear Granger tests, PCMCI variants, and practical implementation via the Tigramite software.
Neural process regression
Wherein an encoder‑decoder architecture is employed to meta‑learn distributions over functions via neural networks that approximate kernel‑like relations, and uncertainty‑aware predictions are produced from context observations for rapid adaptation.
Data storage formats
Wherein are set forth various data storage formats for numeric science, and the columnar Parquet format is singled out for on‑disk columnar organization with efficient compression and analytics use.
Learning from ranking, learning to predict ranking
Learning preferences, ordinal regressions etc
Wherein pairwise preferences are compared using models such as Bradley–Terry, connections to quantile regression are observed, and application to fine‑tuning language models via learning from human preferences is described.
Multi fidelity models
Data-driven multi-scale sampling, multi-resolution, super-resolution
Wherein the practice of combining low- and high-precision models is examined, and the recovery of fine-grained resolution from coarse pixel representations is treated via Gaussian-process inference.
Automatic differentiation
Wherein automatic differentiation is presented as a computational technique, dual‑number and reverse‑mode (backpropagation) formulations are outlined, and applications to ODE sensitivity and neural nets are indicated.
Calibration of probabilistic forecasts
Proper scoring rules, skill scores etc
Wherein the notion of probabilistic calibration is treated as a rule to be enforced, the dictum that eighty‑percent forecasts are mistaken in about twenty percent of cases is adduced, and assessment methods are noted.
Generative industrial design
Wherein diffusion probabilistic fields are applied to the generation of continuous 3D form for object and shape design via neural networks, and domain‑specific denoising architectures are considered.
Learning new languages
Wherein the diglot weave method is described, empirical studies are cited, foreign words are interwoven into native prose, and short‑term vocabulary retention is reported to exceed that of rote drills.
Comfort traps
On edging life with a little friction
Wherein everyday lures — cars, flat‑rate streaming, and easeful households — are depicted as snares that quietly erode capability, and cohousing is proposed as a source of deliberate friction.
Javascript/browser vector graphics
Wherein the browser is enlisted to render vector graphics by libraries and services, and Kroki is noted as a CLI and web service by which varied diagram formats are rendered from simple textual descriptions.
Lotekno
Also, other Sundanese music projects I have been involved in
Wherein a residency in the Sundanese lands of West Java is undertaken, and live electronic jamming is adapted to non‑Western tunings through collaborations with local ensembles and cassette‑based practices.
Playing video
Wherein the mechanics of rendering moving pictures are set forth, and the local transcoding from hard drive to display is delineated, with methods for synchronous remote viewing via named companion apps
Quantization
Wherein the state-space of a system is discretized into representative regions, and its role in coding theory, compression, and mixture-model density approximation is delineated and linked to classification.
Collective care and wellbeing
Wherein communal practices for shared wellbeing are catalogued, and practical rituals such as coordinated meal‑sharing and restorative justice methods are outlined as means for sustaining group care.
Non-negative matrix factorisation
Wherein the decomposition of an element-wise nonnegative matrix into two smaller nonnegative factors is treated as a canonical inverse problem, and its links to optimal transport are noted and sketched.
Belief propagation with loops
Bethe approximation, Kikuchi approximations, loop calculus
Wherein local and global information flows in inference are examined, the Bethe and regional approximations are contrasted, and loop calculus is employed to account for corrections from cycles in factor graphs.
Learning under distribution shift
Wherein the challenge of learning under distribution shift is laid out, and the distinction between hierarchical Bayesian pooling and neural‑network bi‑level/adversarial approaches for Australia–India dataset splits is examined.
Numerical PDE solvers
Wherein a catalogue of open-source numerical PDE solvers is presented, and it is observed that many offerings favour CPU-based finite-element and spectral methods, with GPU support frequently absent.
Efficient food
Economics and resource-efficiency of food production
Wherein the resilience of global food systems and the prospects for resource‑efficient alternatives, such as alternative proteins, and the implications of disaster‑era distribution failures are examined.
Time management
The psychology of getting stuff done
Opt-in self-behaviour control, by my past self, of my current self who is bored and wants to go on lunch break.
Physics-informed neural networks
Wherein a solution is represented by a neural network over coordinates, the PDE residual is penalized via automatic differentiation, and latent stochasticity is handled by polynomial chaos expansions.
Interpersonal relationships
Applied. Love, friendship, trust and oxytocin, considered dyadically
Wherein the modes of cultivating durable attachment and of co-mingling one’s soul with another are presented, and practical methods for managing disputes and deriving mutual value are outlined.
Optimal conditioning
Wherein the problem of selecting conditioning variables for optimal prediction is treated, and the role of learned features in transformers is examined, with compressibility is proposed as an adjunct.
Probability
Wherein an alternative foundation of probability is presented, and its derivation is traced to epistemic concerns about belief and inference rather than to aleatoric notions of chance.
Explainability of human ethical algorithms
Wherein human mechanisms of moral judgment are treated as black boxes to be probed, and influence functions that trace which past examples and perceptual features shape ethical decisions are examined.
Recipes
Wherein the rise of metacookbooks is chronicled, and specific innovations are noted, including salt-cured egg yolks and a no-knead pan pizza, as practical angles on culinary instruction.
Decoupling the economy from energy
Wherein is examined the prospect of economic decoupling from energy through virtual worlds and enclosure of the intellectual commons, with focus on licensed experiences and metaverse residency.
Material basis of the economy
Wherein the material basis of the economy is traced through energy flows, exergy accounting, and global shipping, and the paradox of nations both exporting and importing cars is observed.
Brain-like neuronal computation
Wherein a neuromorphic system is described, in which spiking neurons, Hebbian plasticity, and random synapses are assembled to emulate language acquisition and one‑shot learning at the scale of tens of millions of neurons without backpropagation.
Whom to live amongst
Designing housesholds, bash’es, families, villages
Wherein alternatives to the nuclear family are examined, co‑living practices such as cohousing, living‑apart‑together, and alloparenting are described, and Australian tax incentives for two‑adult units are noted.
DNS
On asking strangers for directions
Wherein the routes my devices take are adjudicated by DNS, whereupon the choice of resolver is noted to affect privacy and security, and encrypted queries to DNS‑over‑TLS servers such as 1.1.1.1 are recommended.
Low-rank matrices
Wherein low-rank matrices are presented as ZZ^H and their Moore–Penrose pseudo-inverse is constructed via the SVD of Z, and a Frobenius-distance formula between UU^H and RR^H is derived.
Matrix inverses
Wherein generalized inverses of singular matrices are enumerated, the Moore–Penrose pseudoinverse is constructed via SVD, the Drazin inverse for matrix index is treated, and updating formulae are noted.
Misrule
Wherein the Feast of Fools aesthetic and DIY subversion are examined, mock chivalric orders and temporary autonomous zones are surveyed, and instances of antisocial punishment in weak‑rule societies are noted.
Weaponised design
Wherein urban aesthetics are described as being repurposed for civic control: planters are installed to obscure unhoused residents, uniforms are catalogued, and graphic design is traced to political ends.
Ecomodernism
Wherein the case for nuclear power is examined as a contested element of technological environmentalism, and urban density is considered alongside greening in pursuit of dematerialized growth.
Coffee and caffeine
Wherein the Yemeni origins of the beverage are noted and the timing of caffeine’s effects on waking cortisol and drug interactions are surveyed, while modest brewing kit and espresso machines are described.
Data dashboards and ML demos
On assuring the client that you are doing something data-sciency because it looks like in the movies
Wherein dashboards are surveyed and the suggestion is made that they be by default configured to withhold half of supplied data as a verification reserve for exploratory analysis, and examples of notebook-to-app frameworks are catalogued.
Bayesian inference for misspecified models
Wherein Gibbs posteriors and power‑likelihood adjustments are examined as remedies for M‑open misspecification, and Bayesian updating is presented as conditional on modelling assumptions.
Taking notes
Wherein various note-taking systems are surveyed and a plain-text blog workflow using VS Codium, Markdown Zettelkasten techniques, and considerations of syncing and end-to-end encryption are noted.
Differential geometry, geometric algebra etc
Wherein a pragmatic guide to manifold wrangling is presented, Clifford algebra and differential‑forms formalisms are compared, and computational tooling plus curated references are provided.
Exit/voice dilemmas
Fix this one or build another?
Wherein the dynamics of categorical departure and restrained protest are examined, and the selective exit of gender‑nonconforming individuals from recorded categories is shown to reshape institutional legibility.
Cohousing in Australia
Plus other low-cost, low-tedium options for secure habitation
Wherein the prospects of cohousing in Australia are surveyed in terse fashion, and coastal flood maps with inner‑west elevation checks are presented as a concrete factor in site selection, and governance and finance options are catalogued.
Emoji
Wherein emoji input methods, shortcodes, pickers and AI mutators are catalogued, and the YAML 1.1 prohibition on emoji in plain scalars is noted, with platform tips for Ubuntu and macOS.
Statistics and machine learning
Wherein a distinction between exploratory and descriptive statistics is drawn, the role of statistics in guarding against marginally viable research is invoked; harmonisation of statistics and machine learning is noted.
Footwear
Wherein zero-drop office shoes and compact, pocketable sandals such as the Brazil-only Sandália Goóc are noted, and spare folding sandals are carried for relief when shoes are soaked.
Where next Sydney freaks?
Wherein it is observed that Marrickville’s bohemian cohort is being priced out by rising rents, and a programme of relocation to cheaper, well‑served suburbs such as Hurstville or Bankstown is proposed.
Random number generation
Wherein a practical pseudo‑RNG implementation is described, the absence of seedable generators in JavaScript is noted, and methods for producing non‑uniform variates from uniform draws are considered.
Informations
Entropies and other measures of surprise
Wherein various formal measures of information are surveyed, and the central role of the logarithm in entropy formulations is emphasized, with Kullback–Leibler divergence and Rényi variants considered.
Academic writing workflow
Cranking the paper mill in the 21st century
Wherein an academic writing workflow is outlined and concrete tools are described, with collaborative LaTeX options and reproducible notebooks such as Jupyter and rticles being recommended.
Autism/allism spectrum
Wherein the satirical term allism is introduced and the workplace implications and the Double Empathy theory, noting that autistic–autistic communication is found to be more effective, are discussed.
Bregman divergences
Wherein a measure of separation between points is presented, being defined from a strictly convex function, exemplified by the squared Euclidean distance, and being employed in mirror descent.
Mirror descent
Wherein the optimization scheme is presented as a dual-space gradient step followed by a projection via a Bregman divergence, an entropic variant for the simplex is exhibited, and efficiency bounds mildly dependent on dimension are given.
Monte Carlo methods
Wherein integrals are approximated by clever sampling, and Bayesian inference via Markov chain samplers and adaptive importance‑sampling is presented, with quasi‑Monte‑Carlo alternatives and convergence rates discussed.
Potential theory in probability
Something about harmonic functions or whatever
Wherein the role of harmonic functions and expectations in Markov processes is examined, and the Dirichlet problem of recovering interior values from boundary data is connected to probabilistic methods.
Gradient descent
Wherein the mechanics of iterative minimisation are presented, the role of momentum is delineated with mention of Nesterov acceleration, and discrete gradient updates are cast as continuous ODE approximations.
Reshipping
Wherein the purchase of foreign goods is impeded by vendors who refuse to ship to Australia, and reshipping services are compared by price and speed, with courier-only options being ruinously costly.
Approximate matrix factorisations and decompositions
Wherein approximate matrix factorisations are surveyed and low-rank-plus-sparse decompositions, randomised sketching and Lanczos approximations are presented as practical routes to scalable matrix approximation.
Browser machine learning
Wherein browser ML tools and runtimes are surveyed, and WebNN, ONNX/WebAssembly, and TVM/MLC efforts for WebGPU compilation and in‑browser execution of models such as Stable Diffusion are described.
Remix and copyright
Wherein the customary borrowing of tunes is chronicled, sampling disputes and the monetization of royalties are set forth, and the prospect of AI-generated mimicry is considered.
Sydney lifestyle theme village
The sunniest mortgage farm in the Southern Hemisphere!
Wherein Sydney is surveyed for its rising rental precarity, suburbs are mapped, sauna culture is noted, millennial out‑migration is traced, and civic projects are listed as possible mitigations.
Sleep
Wherein the habits and mechanisms of sleep are examined, and practical interventions such as blue‑light timing, melatonin dosing, the glymphatic clearance hypothesis, and positional anti‑snoring garments are described.
Virtual machines for curmudgeons
On pretending to have hardware using software
Wherein the varieties of virtualization are surveyed and the libvirt/QEMU‑KVM stack is noted as often simpler and faster on Linux hosts, Apple‑Silicon UTM is mentioned for Mac, and shared‑folder and permission pitfalls are described.
Edge ML
Putting intelligence on chips small enough to get into disconcerting places
Wherein the practice of running compact, quantized neural models on microcontrollers and single‑board computers is described, and deployment challenges such as low‑precision arithmetic and toolchains like TensorFlow Lite and ONNX are adduced.
Naming things
Hashes, UUIDs, haecceities, deep and inscrutable singular Names
Wherein identifier schemes are catalogued and contrasted, and ULID’s lexicographic sortability and proquint’s pronounceability are exhibited as practical criteria for human-facing IDs.
Generalised autoregressive processes
Wherein generalisations of autoregressive processes are presented, the linear AR(1) special case is examined, and variance, correlation and conditional distribution properties are derived.
Secure Scuttlebutt et al
Wherein a peer-to-peer mesh is described, and message logs are copy-pasted between devices over local WiFi, enabling offline-first social feeds via Manyverse, and rooms are used to connect clients through always-on nodes.
Non-uniform signal sampling
Discrete sample representation of continuous signals without a grid
Wherein the problem of non-uniform sampling is examined under Gaussian-process priors, and computational devices such as the Non-uniform FFT and Lomb–Scargle periodogram are adduced for spectral analysis.
Doing it yourself
Peasantpunk, artisanal poverty, hobbyism in the open, handicraft fandoms
Wherein DIY is examined as frugal improvisation, community barnraising, and status signalling, and the practice is linked to jugaad, gift economies, and cottagecore aesthetics.
Firefox
Wherein Firefox is described as a functional web browser, and its containers feature is explained, allowing multiple isolated browsing identities within a single profile to be maintained.
The social brain
Wherein reasoning is described as having been evolved to persuade others rather than to ascertain facts, as exemplified by card‑turning logic tests and models that treat the self as an inner social group.
Genetic programming
Wherein a nature-inspired method is recounted, in which programs are evolved by selection and recombination, is applied notably to symbolic regression problems, and is examined with historical and theoretical notes.
Learnable coarse-graining
Approximate meso-scale physics
Wherein machine-learning approaches are employed to infer coarse-grained force fields from molecular simulation data, including many-body nonpairwise interactions, while computational-scaling and transferability trade-offs are examined.
Plotting in Python
Jack of all trades, old master of none
Wherein Python plotting is surveyed and the tension between interactive web tools and publication-quality output is examined, with Datashader and HoloViz noted for large-data interactivity and Matplotlib’s maze of margins is reported.
AV controller interfaces
Wherein AV controller interfaces are surveyed and the input side is described, including gesture and sensor inputs, OSCQuery auto‑generated web GUIs, device discovery via libmapper, and Kinect calibration using Rulr.
Ergonomics
Wherein standing desks are favored and open‑plan speech intelligibility is observed to impair concentration, while the cathedral‑effect on creativity is treated with scepticism and limited empirical support.
Noise pollution
Wherein inexpensive measures for soundproofing single‑pane glass are described, and musicians’ earplugs for gigs, which preserve tonal balance better than foam plugs, are noted.
Crisis, collapse etc
All your base are belong to dust
Wherein the manner of civilisations' final departures is examined, and the Greenland Norse disappearance is presented as being linked to compounded stressors such as collapsing trade, climatic cooling, and water scarcity.
Polynomial bases
Wherein a survey of polynomial bases is presented and the three-term recurrence for orthogonal polynomials is exhibited, with classical families tied to their weight distributions, e.g. Legendre and Hermite being enumerated.
Gaussian process regression
And classification. And extensions.
Wherein Gaussian process regression is presented as the conditioning of a Gaussian field on observed points to produce a posterior over functions, and is noted to be applied to spatial kriging.
Python spatial statistics
Wherein Python spatial statistics tools are surveyed and a practical installation angle is provided: GDAL is noted as burdensome, and Rasterio, Fiona, and Geostack options that avoid or optionalize GDAL are described.
Semidefinite proramming
Wherein the generalised theory of convex programming is presented, and optimization problems constrained by positive semidefinite matrix inequalities are shown to be expressible and solvable via specialized numerical solvers.
Model averaging, model stacking, model ensembling
On keeping many incorrect hypotheses and using them all as one goodish one
Wherein model averaging, stacking and ensembling are surveyed, a Bayesian stacking variant is contrasted with frequentist averaging, and Jensen’s inequality is invoked to explain gains for convex losses.
Matrix calculus
Wherein matrix-valued functions and arguments are treated, matrix differentials and indexed tensor formalisms are presented, and practical recipes for Jacobians and autodiff in gradient-based algorithms are given.
Healthcare in Australia
Wherein tax incentives for private health insurance are outlined, the time cost of parsing fine print is recorded, a citizen’s perspective is noted, and international comparisons to Switzerland and Indonesia are invoked.
The likelihood principle
Wherein the equivalence of experiments is established when their likelihood functions are proportional, and stopping rules are shown to be immaterial for inference under a correctly specified model.
(Reproducing) kernel tricks
Wherein reproducing kernels in machine learning are examined, and the computational burden of N×N Gram matrices for large datasets is described, with approximate solvers and GPU tools noted.
Devcontainers
Wherein a workflow for using containers as full‑featured development environments is described, an R teaching example is presented, and editor integrations such as VS Code support are noted.
Geoscience
Wherein tools for modeling Earth’s surface and planetary-scale rocky processes are presented using Landlab and Geostack, and software for high-performance geospatial processing and generation of random 3D geological models is cited.
Bio marker tracking
Instrumenting health to improve it
Wherein monthly serial blood panels and genomic markers are recorded and analyzed using personal-data tooling, and are used to inform quotidian fitness and personalised preventative regimens.
Hypothesis tests, statistical
Wherein hypothesis tests are presented as decision tools for A/B experiments, and classical null tests are described as linear‑model checks with p‑values, power and sample‑size considerations
3d data
A grab bag of point clouds, volumetric data and photogrammetry
Wherein 3D datasets and software are catalogued, with photogrammetry workflows and neural‑implicit representations being treated, and Objaverse’s 800K+ annotated models being recorded as a concrete resource.
Codebraid
Minimalist scientific workbook
Wherein a scientific workbook is presented, driven by Jupyter kernels and Pandoc, and wherein multiple programming languages with inline execution, isolated sessions, and minimal diffs for version control are permitted.
Calendars and scheduling software
Wherein the mechanics of synchronising calendars and contacts via CalDAV and CardDAV are examined, and the practical friction of iCloud compatibility and server choice for desktop and mobile clients is described.
Gaussian process ensembles
Bayesian committee machines, product-of-experts
Wherein Gaussian process ensembles are constructed from committees of observation subsets, and regression is enabled to be scaled through distributed, nested aggregation of local predictors.
Synchronising config files across machines
Wherein dotfiles are advised to be managed via a bare git repository or by initializing a git repo in $HOME, and mackup’s fragility with non‑ASCII filenames is reported.
Blogdown
Plus other RMarkdown-derived scholarly blogging systems
Wherein the academic blogging system is described and its interplay with R Markdown and Hugo is examined, including handling of citations, equations, and a noted need to remove stale .lock~ files to resume builds.
The returns on hierarchy in group coordination
Wherein the efficiency gains of hierarchical coordination are weighed against unequal status costs to lower ranks, and attention inequalities are noted via Lorenz-curve–style concentration.
Homebrew
Minimalist software management for POSIX systems
Wherein a userland package manager is presented for procuring codecs and content-related tools on Linux and macOS, and the presence of a macOS .pkg installer since version four is recorded.
Model-based NN
Upproximateing unrolled algorithms
Wherein learning is guided by prior domain knowledge and iterative algorithms are unrolled into network layers, enabling interpretable, parameter‑sparse models that are trained from limited data.
Position encoding
Wherein position is treated as an input to neural architectures, and sine and cosine Fourier features are described as a common concrete encoding, being used in transformers, PINNs, and implicit representation nets.
Journalism, normative
Wherein the incentives undermining reporting are sketched and mechanism‑design remedies are proposed, including prediction‑market funding and cross‑partisan reward rules for verification and prioritisation of news.
Decaying sinusoid dictionaries
Wherein analytic inner products and normalization constants for decaying sinusoid atoms are derived using a top‑hat window, and matching pursuit fitting of autocorrelograms is described.
Markdown
An itemised list of the esoteric difficulties involved in bullet points
Wherein the choice of markdown for research composition is narrated, and pandoc-compatible tooling, HTML-to-markdown converters and MyST extensions for academic publishing are delineated.
Spatial data in R
Wherein R is employed as a geographic information system, and the modern simple‑features sf workflow with GeoJSON export is described for producing maps, tiles, and interactive web integration.
Noise contrastive estimation
Wherein a model is trained to distinguish observed samples from a chosen noise distribution via a binary classifier, and an unnormalized data density is thereby estimated through a learned likelihood ratio.
Bayesian model calibration
Wherein surrogate optimisation via Gaussian process emulators and adaptive design of experiments are employed to infer parameters from simulation-heavy models, and maximum mean discrepancy is used as a fitting criterion.
Climate change
Wherein climate simulations, regional impacts, economics and mitigation are surveyed, and both 1,200 years of Kyoto cherry‑blossom records and a $571 billion projected property‑market loss by 2030 are cited.
Institutions and governance for mass conversation
Discourse red in tooth and claw versus the sovereign
Discussion of hate-speech, various internet censorship flashpoints including for example sexual violence; links to critics of speech norms from diverse places on the…
Matrix norms, divergences, metrics
Wherein singular values and Schatten, Frobenius, and nuclear norms are surveyed, operator and Frobenius norms are compared via singular-value relations and inequalities, and trace formulations are exhibited.
Stylus input
Wherein a survey of stylus devices is presented and the author’s Huion Kamvas 13‑inch model is noted to require a 3‑in‑1 cable under macOS, thereby occupying an HDMI port.
Bayes functional regression
Wherein various Bayesian methods are surveyed and Gaussian process regression is presented as a junction for estimands that are functions over continuous domains, with manifold and SDE connections indicated.
Non-Gaussian Bayesian functional regression
Wherein an investigation is presented into regression with non-Gaussian random fields, attention being paid to higher moments and the use of sparse stochastic process priors as a practical alternative
State filtering for hidden Markov models
Wherein state estimation for hidden Markov models is treated as recursive, online Bayesian filtering, and the Kalman–Bucy and particle-filter variants are presented for linear and nonlinear dynamics
Nerdview
Wherein the phenomenon of specialised jargon is observed in public communications, and NSW tram warning signage and a government travel agent’s baffling help pages are documented, noting a QBT→CMT name change.
Neural learning dynamical systems
Wherein neural methods for learning dynamical systems are described, emphasis is placed on stochastic extensions such as SDEs and jump processes, and issues of stability and recursive identification are considered.
Dropbox if you must
Wherein the use of rclone is described to synchronise a Dropbox folder with a local git repository, and client‑free FUSE mounts, sandboxing and on‑the‑fly encryption are presented as mitigations.
Saying “Bayes” is not enough
The other secret steps to doing Bayesian statistics
Wherein Bayesian methods are acknowledged but their insufficiency is asserted due to heavy mathematical and computational burdens, and because priors and hypothesis generation are not provided by Bayes.
Gradient descent, first-order, stochastic
Wherein it is shown that noisy first‑order information is used to minimize objectives in very high dimensions, and martingale arguments, continuous‑limit SDEs, and variance‑reduction techniques are invoked.
Belief propagation and related algorithms
Wherein belief propagation is presented as message passing on tree factor graphs, and the marginalisation integrals that define factor-to-variable messages are exhibited, with Gaussian variants being noted.
Canonical correlation
Wherein canonical correlation is presented through singular value decomposition, an SVD-based derivation is supplied, and classical statistical tests are revealed as linear models judging whether coefficients are nonzero.
Method of Adjoints for differentiating through ODEs
Wherein the adjoint technique is expounded, and a backward PDE is constructed to furnish gradients of the forward time‑evolving system, such gradients being used to differentiate likelihoods via automatic differentiation.
IPython
Wherein the interactive Python shell is described, and its rich-display protocol for inline graphics is explained, while output caching and debugger invocation via set_trace() are detailed.
Matrix square roots
Whitening, preconditioning etc
Wherein matrix square roots are treated as either X X = M or X Xᵗ = M, and formulas are given for (αI+UVᵂ)^{1/2} via k×k matrices and a Denman–Beaver iteration is described.
Conformal prediction
Wherein distribution-free predictive intervals for arbitrary machine learning models are constructed via a conformity function, their empirical coverage properties are assessed, and robustness under dataset shift is examined.
Multi-objective optimisation
Wherein the predicament of unknown weights in weighted‑sum formulations is examined, the Pareto‑front nature of hyperparameter tuning is evinced, and recourse to Lagrange multipliers is noted.
Optimal transport inference
Wherein optimal-transport based inference is surveyed, and practical expedients such as the estimation of Monge transport maps or minibatched Sinkhorn approximations for Wasserstein losses are indicated.
Distances between Gaussian distributions
Nearly equivalent to distances between symmetric positive definite matrices
Wherein the relations between Gaussian laws are examined by explicit formulas for distances such as Wasserstein‑2, Kullback–Leibler and Hellinger, and the role of covariance matrix geometry and matrix‑logarithm geodesics is expounded.
Reparameterization methods for MC gradient estimation
Wherein the reparameterization trick is presented as a transformation of noise via a differentiable map so that Monte Carlo pathwise gradients are computed, as used in variational autoencoders and normalising flows.
Combining kernels
Wherein kernels are treated as Gaussian-process covariance functions, and it is shown that sums, products and warping are preserved as kernels, with induced feature maps concatenated rather than summed.
Numerical libraries
Wherein a bewildering multitude of numerical libraries is surveyed, and XLA-backed Python ecosystems, LAPACK descendants and Rust crates that compile to WebAssembly are noted, with vendor stacks found bulky.
Neural implicit representations
Neural nets as coordinate mappings
Wherein neural signals are parameterized as continuous functions mapping coordinates to values, permitting sampling at arbitrary resolution and enabling signed‑distance level‑set representations for 3D shapes.
Neural rendering
Wherein Gaussian splatting and neural radiance fields are surveyed, and implicit neural functions are employed to represent 3D scenes as coordinate-to-color-and-density maps, and practical reconstruction from images is discussed.
Covariance estimation
Wherein the estimation of covariance matrices is considered with attention to large‑p inversion difficulties, and Bayesian inverse‑Wishart priors and sandwich estimators are described.
Generative AI workflows and hacks 2023
Wherein a compendium of ephemeral notes and links on prompt engineering, model‑coding, and the running of large models in browsers is presented, and a continued habit into 2024 is indicated.
Lua
Wherein an embeddable Brazilian scripting tongue is described, its LuaJIT acceleration and use inside the document tool pandoc are noted, and its role as unobtrusive glue for audio, games, and typesetting is indicated.
Model order reduction
Wherein the approximation of complex models by projection onto a dominant subspace is presented, the practice of emulation via low-dimensional summaries is surveyed, and the prevalence of specialized terminology is noted.
ΦFlow
A modern python computational fluid dynamics library for ML research
Wherein a differentiable PDE framework is set forth, and tight integration with PyTorch, Jax, and TensorFlow is provided, enabling GPU‑accelerated, fully differentiable fluid simulations, while a web interface for live visualisations is offered.
Code editors
The best thing since punchcards
Wherein editor loyalties are chronicled and the tendency for hackable editors to be enveloped into general OS interfaces is observed, and Visual Studio Code is noted as the present preference.
Performance indicators, Measurement, analytics
Wherein the principles of dashboards and analytics are set forth, and the use of layered tooling such as the Kedro Python framework for organizing data work for exploratory analysis and measurement is described.
Online collaboration tools
Wherein alternatives to Google Docs for collaborative editing are catalogued, including privacy-first CRDT platforms such as AFFiNE, zero-knowledge CryptPad, and the collaborative LaTeX editor Overleaf.
Leakage in predictive models
Wherein cross-validation is shown to permit inadvertent access to future information through procedures like target encoding and stacking, causing model evaluation to be compromised and hold-out testing to be recommended.
Cheap talk
Sapir-Whorf politeness, virtue signalling etc
Wherein the costs of ostensibly cheap talk are catalogued, and the role of gestures such as gender-neutral job descriptions and Acknowledgement of Country is examined as institutional signals affecting common knowledge and ritual.
Optimal rotations
Wherein an optimal rotation is treated as a linear whitening transform, and an orthonormal matrix is found by minimising a trace norm via matrix calculus, as in algorithmic practice
Webmail clients
Wherein a survey of self-hosted webmail clients is presented, and single-user operation via local Maildir with Notmuch indexing is described as a practical alternative to desktop Linux mailers.
Expectation propagation
Wherein expectation propagation is presented as a message‑passing scheme in which local factors are replaced by approximations via moment matching, tilted distributions are formed, and KL(p‖q) is minimised.
MLP-Mixer neural networks
Wherein position encoding is employed and interleaved token- and channel-wise multilayer perceptrons are used, and image structure is handled without convolutional kernels, is expounded.
Contemporary neo-feudalism & endimming
Wherein the emergence of neo-feudal orders is described, clientelism and patronage are mobilised as the engine of governance, and online troll-cultures are invoked as ideological promoters.
Practical text generation and writing assistants
Wherein the rise of consumer-facing writing assistants is chronicled and the ubiquity of spammy tools, the need for smooth user interfaces for honest writers, and a Galactica controversy are noted.
PDF viewers
On turning texts into font rendering errors
Wherein a desire for shared annotations across Linux, macOS, iOS, and Android is considered, and SyncTeX support, annotation-exchange limitations, and inability to overwrite PDFs in some viewers are noted.
(Nearly-)Convex relation of nonconvex problems
Wherein a least-squares relaxation is examined, and an overparameterization is shown to produce analytically tractable formulations that are employed in kernel methods, sparse coding, and matrix factorization.
Markdown editors
Wherein markdown editors are surveyed and a reliance on tools offering mathematics preview and document syncing is recorded, and both GUI applications and command-line viewers are catalogued for comparison
Sparse coding with learnable dictionaries
Wherein adaptive dictionaries for sparse coding are presented and methods for batch, online, and shift‑invariant transform learning are surveyed, and large‑scale learning and invariance issues are addressed.
Online whiteboards
Wherein online whiteboards are surveyed and the need for stylus input for writing-heavy mathematics is noted; an open-source Excalidraw with end-to-end encryption is listed among candidates, and commercial alternatives with tiered pricing are catalogued.
Tensor decompositions
Wherein the theory of multidimensional arrays is introduced and three popular formats—CANDECOMP/PARAFAC, Tucker, and Tensor Train—are presented, and software libraries for experimentation are noted.
Transport maps
Inference by measure transport, low-dimensional coupling…
Wherein probability measures are mapped into one another by explicit transports, connections to normalising flows are noted, and cost-weighted preferences are introduced, producing optimal transport metrics.
UX
Wherein the study of user experience is presented as an examination of interface interruptions and prescriptive design rules, and human attention is treated as a constrained resource to be managed.
Science Fiction
Wherein the author’s reading habits are chronicled: preference for political‑economy plots, explosive escapism, and philosophically troubling tales is set forth, with audiobooks noted as a practical filter.
Disgust
Wherein the sentiment of disgust is examined as a correlate of political conservatism and pathogen-avoidance, and interpersonal contamination sensitivity is identified as most predictive of social conservatism.
External validity
When does what I learn on one data set apply to another?
Wherein external validity is examined as the transferability of learned models, and it is noted that feedback from deployed algorithms, as in traffic routing, is capable of altering the environment in which they are applied.
The Matrix-Gaussian distribution
Wherein the Kronecker-structured covariance of rows and columns is exhibited and the density is expressed via a trace form, and computational savings for sampling and pdf evaluation are obtained by avoiding full vectorization.
Interoperating with R
Wherein R is shown to be bridged to compiled languages and Python, and tools such as Rcpp, JuliaCall and TMB are presented, while an HDF5 filesystem workaround is noted for large sparse-matrix transfer.
Jupyter UI wrangling
What happens inside those notebook cells doesnt stay inside tgose notebook cells
Wherein the mechanics of the Jupyter user interface are laid out in sober terms, and the procedure for exporting notebooks as reveal.js slides is demonstrated alongside notes on rich display and widgets.
UIs in Python
interacting with an app, a python app, without too much dicking about
Wherein stream-processing and web‑engine dashboards for live GUIs are surveyed, and a catalogue of toolkits from tkinter and wxPython to Qt, Kivy and web‑backed dashboards is presented.
Generative flow nets
Wherein generative flow nets are described as being trained to sample candidates in proportion to a reward, whereby costly MCMC work is amortized, partition functions and marginals are estimated, and distributions over sets and graphs are represented.
Shells
Wherein the complex history of Unix shells is surveyed and the persistence of archaic quirks is noted, including the legacy problem of filenames with spaces traced to 1979 and varied modern alternatives.
Last-layer Bayes neural nets
Bayesian and other probabilistic inference in overparameterized ML
Wherein the terminal layer is treated as a linear model with Gaussian i.i.d. errors, its weights are inferred via pseudo‑inverse least‑squares, and a statistical justification is supplied via maximum‑likelihood
Last-layer Bayes neural nets
Bayesian and other probabilistic inference in overparameterized ML
Wherein the network is reduced to a deterministic feature extractor and the final weights are treated as a Bayesian linear model, so predictive densities are derived via a Laplace-style closed-form approximation.
IDEs for Julia
Wherein IDEs and workflows for Julia are surveyed in a matter‑of‑fact manner, Revise.jl and Pluto are cited, and steps are given for preventing duplicate Jupyter installs when using IJulia with Conda.
manim
pedagogic animations via python
Wherein a Python library for mathematical animation is treated as a passion project of Grant Sanderson and is documented with practical install tips, SVG sizing quirks, and notes on an experimental OpenGL renderer.
Visualising probabilistic graphical models
Also related models, such as Neural nets
The diagrams I most often need are directed flow graphs, a formal mathematical cousin of the flowchart, which can represent graphical models and neural nets. That is, my…
Maths hacks
Wherein classical inequalities and integration tricks are presented, including Young’s inequality for products and a Maclaurin integration formula that yields an antiderivative for e^{x^2} without termwise power series
Web API automation
Wherein web services are automated via published APIs and workflow platforms, and alternative self‑hosted tools such as n8n and Huginn are noted, demonstrating integrations without using a browser.
When to argue ad hominem
Wherein the calculus of ad hominem is considered and instances when base rates and recursive credibility checks are used to justify personal-attribute rebuttals are delineated.
Conference posters
Wherein the conference poster is treated as the least regarded of three academic rites, and practical guidance on layout, Scribus LaTeX hacks and poster-sized plotting is supplied.
Web scraping
Turns out we can get information off the internet
Wherein web pages are parsed for structured data, parsed outputs are converted into RSS feeds by configured parsers, and deployments are orchestrated across cloud services to run the extraction at scale.
Browser graphics
Wherein browser graphics are cataloged and the use of WebGL (OpenGL ES) for 3D rendering, including shader-based effects and lens flare, is described in succinct technical terms.
Statistics and ML in Python
Wherein Python is presented as a statistical and ML workhorse, its matrix- and DataFrame-style ecosystems are surveyed, and practical interoperation with R via Feather/Arrow files is demonstrated.
Javascript audio
Every program will expand until the point that it can generate cheezy techno
Wherein web audio in the browser is examined and the lack of a native white-noise generator and scheduling primitives is described, libraries and examples are catalogued, and MIDI support is noted.
Research data sharing
Wherein the practices of sharing research data are considered, and the choice between static DOI-backed repositories such as Zenodo and incrementally updated, versioned collaborative stores for ongoing experiments is examined.
Matplotlib
A way to draw things in python, which is better than no way to draw things in python
Wherein matplotlib is presented as the classic Python plotting library, its Axes/Axis terminology is clarified and its role as a backend for wrappers such as seaborn and proplot is noted.
Are they too old/young for me?
Dating, gender, age, and equity
Wherein the question of being too old or young is examined and an Australian population-percentile method is proposed, using ABS 2021 age data to match partners by equivalent gender-specific percentile.
Bounded rationality at large
It is as if we knew what we were doing
Wherein markets are examined as mechanisms by which prices are found despite aggregate ignorance and boundedly rational agents, via noisy short-sighted trades, and simple measures of collective rationality are proposed for institutions.
Drugs, recreational
Wherein the social, legal and therapeutic contours of nonmedical psychoactive substances are surveyed, and emerging psychedelic therapies and practical MDMA harm‑reduction measures are noted.
Gradients and message-passing
Cleaving reality at the joint, then summing it at the marginal
Wherein automatic differentiation is presented as message-passing, and Bayes-by-backprop is linked to variational message-passing through the chain rule, with stochastic updates being noted.
Comfy Ubuntu
Wherein Ubuntu’s bloated defaults are presented as a pragmatic base for HOWTOs, the ambiguity of snap/flatpak/deb packaging is noted, and alternatives such as Pop!_OS and Homebrew installations are surveyed.
Highly performative computing
On getting things done on the Big Computer your supervisor is convinced will solve that Big Problem with Big Data because that was the phrasing used in the previous funding…
Password management
Don’t re-use passwords for different services; that would be foolish. Don’t try to remember many passwords; that would be hard. Use a password manager, which remembers many…
Image search
Wherein image search by visual similarity is described, and a CPU‑only dataset curation tool that scales to millions of images is noted, while desktop and web reverse‑search options are listed for finding duplicates, anomalies, and near‑matches.
Special LaTeX symbols
Wherein the handling of special LaTeX symbols is treated, and methods for inserting emoji via Unicode fonts or packages and the 18,150‑entry Comprehensive LaTeX Symbol List are described.
Elliptical distributions
Wherein a density is expressed via a scalar function g of the Mahalanobis quadratic form (x−μ)'Σ^{-1}(x−μ), Σ is noted to be proportional to the covariance when it exists, and many robust M‑estimator losses are encompassed.
BibLaTeX
Wherein BibLaTeX is presented as a LaTeX bibliography system, was paired with the biber backend for full Unicode support, and was shown to render bibliographies with a dedicated print command and support compound inline citations.
Economic development
Wherein institutional arrangements are examined, the persistence of nepotistic kleptocracies is described, and the institutional pathways from poverty to high-functioning states are outlined with attention to governance reforms.
Fitness
Getting swole and/or deferring death
Wherein the author’s bodyweight training regime is described, guidance on strength, flexibility, and protein timing is surveyed, and workout apps such as Fitbod are noted for planning and tracking.
Synchronising files across machines
Wherein peer-to-peer and host-based tools are surveyed and a recommendation of Syncthing for local peer syncing and rclone encryption-backed bridges to cloud storage is presented.
Open Source (mostly software)
On published schematics for cheap objects
Wherein the sociology and logistics of community-developed code are examined, with maintainer time burdens, upstream-first patching practices, funding channels such as BackYourStack, and contributor-licensing trade-offs being outlined.
COVID-19 in practice
Wherein the limits of rapid antigen tests for Omicron are examined, with saliva PCR cycle-threshold discordance noted, throat swabbing and CO2 ventilation monitoring are considered, and microlife-style risk calculators are recommended.
Density ratio tricks
Wherein the estimation of distributional likelihood ratios is presented, and classifiers are employed to recover density ratios for use in importance-weighted estimation and two-sample comparison.
Growing up
Wherein the reader is shown how childhood emotional strategies are inherited and persist into adulthood, and the tendency to presume adults conceal truths—as with Santa—is traced as an enduring maladaptation.
Statistical relational learning
Wherein lifted inference techniques are employed to reduce complexity in probabilistic logical models, and exchangeability and projectivity issues for social-graph inference are examined.
Models of computation
Turing machines, λ-calculus, term-rewriting and other models of what may be computed.
Wherein the nature of computation is surveyed and it is shown that mundane systems are rendered Turing-complete, as when the x86 mov instruction is repurposed to implement arbitrary computation.
Conditioning non-specific advice
Wherein advice is treated as a statistical problem, readers' class mismatch is noted, the social world is characterised as adaptive systems, and the search for the optimal amount of X and for actionable alpha is described
Hamiltonian and Langevin Monte Carlo
Physics might be on to something
Wherein the exploration of the typical set by measure-preserving Hamiltonian flows is described, and symplectic integrators, intermittent momentum resampling in NUTS, and reflection/refraction for discontinuities are outlined.
But what can I do?
Recommended behaviour to make society better is to think, then act
Wherein the reader is shown that, in Australia, collective agency is to be exercised by recurring donations, volunteering and community organising to mobilise political leverage for climate adaptation and social resilience.
Starfish problems
Recommended behaviour to make society better is to think, then act
Wherein the marginal impact of a child’s efforts to toss beached starfish back to sea is examined, and the question of whether rescue procedures and best-practice methods could be improved is posed.
Interaction effects and subgroups are probably what we want to estimate
Wherein the necessity of estimating interactions and subgroups is presented, the often-enormous sample sizes (commonly an order of magnitude larger than for main effects) required for detection are noted, and the risks of post hoc selection and confounding in observational data are described.
Personalized medicine
Wherein self-experimentation and DIY genetic testing are recounted, single-subject biomarker tracking is examined, and the interplay of personal narrative and experimental protocols is documented.
How to communicate
The skill of communicating in the highly artificial situations of the modern human. Such crucial skills. Often not taught. Worse, we systematically fail to realize we lack…
Scientific writing
In which tips are given for the projection of status through nominal phrases and passive voice
Wherein the rituals of academese are examined and the tradeoff between clarity and status signalling is described, with attention paid to mathematical notation and editorial incentives.
Blogroll
Contrary opinions
Culture wars
On our the ascendancy of virality over importance
Wherein the racial and sexual skirmishes of public discourse are depicted as a theatrical psychodrama of repenters and repressers, and a market of intellectuals is shown to vend moral performances.
Electric cars
Wherein early electric vehicles are chronicled from 1904, and examples of minibuses and trucks used to supply household lighting and off-grid power are presented in documented detail.
Organising a photo collection
Wherein a survey of tools for sorting, metadata handling, and local AI tagging is presented, including the exiftool command to erase embedded metadata and mention of self‑hosted LibrePhotos.
Feelings, applied
Wherein feelings are examined as loss-and-reward functions and as socially constructed signals, and practical approaches such as acceptance and commitment, labeling, and conversational externalization are described for coexisting with them.
Email hosts
Who handles my mail? And what do they do with it?
Wherein a catalogue of email hosts is presented, and the option of paid host‑proof encryption in favourable jurisdictions is examined as a means to reduce corporate and state collection of personal behaviour data.
Emancipating my tribe
Inclusivism and exclusivism in sacred and secular subcultures. Or, scaling up the in-group.
Wherein the dynamics of collectivist tribes are examined, distinctions between tribal behavior and problem-solving are drawn, and ladders between communities (e.g., 4chan to Less Wrong) are described.
Deep learning as a dynamical system
Wherein neural networks are treated as dynamical systems, ResNets are shown to approximate ODE/PDEs, energy‑conservation constraints are imposed, and layer discretization and ODE‑learning methods are described.
The edge of chaos
Computation, evolution, competition and other past-times of faculty
Wherein the relation between criticality and the trainability of deep neural networks is examined, and the depth-to-width aspect ratio is noted, as a mechanism by which exploding and vanishing gradients are avoided.
Pluralism, multiculturalism, politics
Wherein religious models of pluralism are examined as templates for political coexistence, institutions from liberal democracies to empires are surveyed, and civic norms like anti‑littering are tested.
Casual anthropic principles
Convenience-sampling lived human experience
Wherein observational selection effects are surveyed, and the constraint that the observer‑sampling process imposes on inferred regularities — from multiverses to social filter bubbles — is set forth.
Randomised linear algebra
Wherein randomised matrix projections are surveyed and Hutchinson’s trace estimator and stochastic Lanczos quadrature are presented as tools for cheap trace and log‑determinant approximations, with links to randomised regression.
Online learning
Wherein the framework of online learning is presented and the notion of regret bounds is explained, and an incremental covariance update (Welford’s recurrence) is given as a concrete method.
Anomaly detection
I don’t define what is abnormal, but I know it when I see it
Wherein anomaly detection is considered, with emphasis on high-dimensional and time-series methods, and an ocean-hydrology model outlier shaped like a distended whale is examined.
Neural tangent kernel
Wherein the behavior of infinitely wide feed‑forward networks is examined, and it is shown that their training dynamics are governed by a fixed kernel, thereby rendering gradient descent equivalent to kernel regression.
Editing images
Wherein methods of refining images for online use are disclosed, and browser utilities for compression and anonymisation are cited, with specific mention of Squoosh for compaction and an image scrubber for privacy.
Posterior Gamma process samples by updating prior samples
Wherein a method is proposed for transforming prior Gamma process samples into posterior samples, and it is shown that updates are effected by rate changes and added atoms with summaries given by event counts and cumulative mass.
Continuous and equilibrium probabilistic graphical models
Wherein probabilistic graphical models over continua are considered, and a Gaussian process with continuous covariance kernel is shown to induce local influence regions on the index space R^n.
Tribal sorting and polarization
Social fragmentation by browser cookies
Wherein tribal sorting and polarization are examined, YouTube recommendation bubbles are shown to shape distinct media ecosystems, and partisan outlets are traced to attention‑for‑revenue feedback loops.
Expert Forecasting
Wherein professional forecasters’ biased and autocorrelated errors are attributed to Bayesian agents learning low‑frequency features of the data generating process, as shown for nominal interest rates and GDP
Forecasting
Wherein recursive estimation, the horizon-and-history framing for next-step prediction, and software ecosystems for probabilistic calibration, model selection, and M4 competition benchmarks are presented.
Intellectual property
especially in science and technology.
Wherein the law and commerce of inventions are surveyed, patent trolls, DRM, and open-source tensions are catalogued, and the rapid copying of crowdfunded hardware by Chinese manufacturers before funding completes is noted.
Transcoding
Digital recordings and converting between them.
Wherein the procedures of converting audiovisual formats are set forth, and concrete methods for offline capture and disc ripping are detailed, including youtube-dl, ffmpeg and HandBrake workflows.
Precision matrix estimation
Wherein the inversion of the covariance is treated as a task beset by large p and large n, and iterative schemes such as conjugate gradients, Lanczos and QUIC are presented as practical routes to approximate precision matrices
(Weighted) least squares fits
Wherein weighted and iteratively reweighted least squares are treated, connections to Fourier-domain relaxations are noted, and practical solver tooling such as Ceres and KeOps is indicated.
Biomimetic algorithms
Wherein biomimetic algorithms are presented; particle-swarm heuristics and artificial-chemistry models are invoked as methods for search and optimization in problems lacking clear analytic solutions.
Multi level marketing
Pyramid schemes, Ponzi schemes, Newcomb unboxing
Wherein links between multi‑level marketing, academia, and cryptocurrencies are examined as nondeterministic pyramid schemes, a case study and podcast are cited, and questions about evidential decision theory are raised.
Learning graphical models from data
Also, causal discovery, structure discovery
Wherein methods for inferring independence graphs from data are surveyed, and a continuous‑optimization approach to learning directed acyclic graphs via a differentiable acyclicity constraint is described.
Bayesian inverse problems in function space
a.k.a. Bayesian calibration, model uncertainty for PDEs and other wibbly, blobby things
Wherein the Bayesian treatment of inverse problems in function space is presented, and the distinction between measurement discretization and solution discretization via projection operators is examined for PDE-driven spatiotemporal models.
Squads
Wherein it is observed that loose networks are reframed as committed cohorts directed toward joint projects and a shared sense of self, and Hamming circle practices are invoked as a model for coordinated friendship
Depression
Wherein the author’s lifelong encounters with afflicted intimates are chronicled and practical notes on assisting them, including surprising mention of sleep deprivation as a transient remedy, are laid out.
The interpretation of RV densities as point process intensities and vice versa
Point process of observations ↔ observation of a point process
Wherein the reinterpretation of densities as point-process intensities is presented, a basis-function expansion is employed, and an equivalence is shown up to a scaling of weights so that ω_j = n w_j.
Gaussian process inference by partial updates
Wherein Gaussian process inference is examined by partial updates, subsampling observations and sites is considered, and Wasserstein and Sinkhorn bounds for approximate posteriors are presented.
Generalised Ornstein-Uhlenbeck processes
Wherein Ornstein-Uhlenbeck generalisations are described and are shown to be induced by Lévy-process bridges in discrete and continuous time, with stationarity and covariance specified by a Lyapunov equation.
Penalised/regularised regression
Wherein penalised regression is presented as a remedy for ill‑conditioned inverse problems, and L2 (ridge) penalties are noted for yielding tractable information‑criteria and a Bayesian prior interpretation.
Eye/head-tracking input
Wherein gaze is employed to pilot the mouse cursor, consumer head and eye trackers are characterized as gamer-oriented devices, and their integration with voice-control systems is noted.
Data versioning
Wherein methods for tracking dataset changes are surveyed, and tools for handling large remote assets and provenance (S3-backed stores, git‑annex, DVC, Pachyderm) are enumerated.
Laplace approximations in inference
Lightweight uncertainties, especially for heavy neural nets
Wherein the posterior is approximated by a Gaussian about the MAP, the network Jacobian is used to propagate parameter uncertainty to output variance, and tractable Gaussian predictive densities are obtained.
Gaussian belief propagation
Least squares at maximal elaboration
Wherein Gaussian belief propagation is presented as a message‑passing algorithm on jointly Gaussian variables, and updates are executed as simple linear‑algebra operations in the information (precision) parameterisation.
Scattering transforms
Wherein scattering transforms are presented as constructions that are derived from wavelets and convolutions, and are shown to encode translation and rotation invariance and higher moments of random fields.
Distributional robustness in inference
Wherein distributional mis-specification within a Wasserstein ball is treated, and ties to causal inference, differential privacy, and adversarial learning are recorded.
State space reconstruction
Wherein delay embedding via Takens’ theorem is outlined, and methods for reconstructing hidden state spaces from time series are described, including Takens embeddings, symbolic dynamics, and Hirata’s graph mapping.
Managing people
Making piece with the fact that I cannot do everything myself
Wherein the art of managing people is treated as a practical ledger, and the MOCHA regimen is laid out to require a single Owner and to delineate Manager, Consulted, Helper, and Approver roles.
Containerized apps (for scientists)
Doing things that previously took 0.5 computers using 0.4 computers
Wherein containerized apps are presented as lightweight, reproducible execution environments for science, and Apptainer is noted as an HPC‑oriented alternative to Docker, while inner‑loop development workflows are considered.
Factorial hidden Markov models
Wherein the hidden state is factorized into independent latent chains, and inference is rendered tractable by separable state variables, with an account of compatibility with neural architectures.
Mellin transforms
Wherein the Mellin transform is presented as a scale‑invariant integral transform and is applied to the study of products, reciprocals and powers of random variables, and its role in multiplicative limit laws is noted.
Semantics
Compressed representations of reality for syntactic agents; which might be what meaning means
Wherein the mapping between linguistic tokens and their referents is surveyed, and attention is given to vector grounding in transformers, MRI evidence for shared conceptualisations, and object‑anchored embeddings.
Cluster B personality disorders
Wherein Cluster B personality types are delineated, population prevalence is cited (eg 6.2% NPD, 3.7% ASPD), and their recognition is presented as useful for managing interpersonal relationships.
Generic variance reduction in Monte Carlo samplers
Wherein a survey of Monte Carlo variance reduction methods is presented, and the question of generic applicability is examined through consideration of sample diversity, Rao-Blackwellization, and preliminary references.
Julia interoperation and foreign function interfaces
Wherein Julia is set forth as being bridged to C, Python, and R, and wherein C calls are compelled to have their call signatures specified at invocation time, and external programs are made runnable.
Elliptical belief propagation
Generalized least generalized squares
Wherein the Gaussian assumption is relinquished and Mahalanobis distance is invoked, robust Huber and Student‑t updates are employed, and outlier effects are inferred by adaptive loss scaling.
Apptainer
Containerized apps for research
Wherein the container platform is described, its single-file, cryptographically signable images are noted, and operation without root privileges on HPC, cloud, and laptops is reported.
Recommender systems
Wherein the evolution of techniques and tooling for personalised recommendation is surveyed, and matrix factorisation methods and practical libraries such as Microsoft Recommenders are referenced.
Practical LaTeX fonts and character sets
Wherein the practicalities of LaTeX font and character handling are delineated, and the contrast between pdfLaTeX’s legacy encodings and XeTeX/LuaTeX’s Unicode fonts is noted, with BibLaTeX cited as Unicode-friendly.
Video conferencing
and other ways of working together, or at least around each other
Wherein video conferencing is surveyed as a catalogue of tools and practices, with attention paid to virtual hallways for serendipity and to technical routing of audio to manage meeting spaces.
Automatic differentiation in Julia
Wherein various Julia automatic-differentiation approaches are surveyed, ChainRulesCore-backed backend-agnostic rules are recommended, and differentiating through Fourier interpolation with Hessian tricks is noted.
Neural nets with implicit layers
Wherein neural networks are presented as layers defined by fixed‑point optimisations, and gradients are obtained via the implicit function theorem, with convex optimisation layers exposed as differentiable modules.
Proof assistants
Wherein a particular slant on computational symbolic mathematics for foundational applications is described, and the Lean proof assistant is noted to provide a VS Code interface for formal theorem development.
Observablejs
Wherein is described a browser-hosted scientific workbook built atop D3, whose Observable Plot library is open-source and is used to produce concise, interactive tabular visualizations, with a free hosting option for notebooks.
Vector icons
Wherein various icon libraries are cataloged and methods for converting SVGs to fonts via FontForge and for building bespoke icon fonts with Glyphter or Fontello are described.
Social factors in information security
Our revealed preference for revealing our preferences
Wherein the misalignment of incentives is likened to pollution, and the burdens of secrecy are shown to be shifted onto individuals, with SMS routing and liability rules cited as concrete vectors of risk.
Academic blogging workflow
Wherein the author’s notes are kept as plain text and are published as HTML, and mathematical markup alongside citation management is enabled via static-site tooling such as Hugo and pandoc.
Instumental variables and two stage regression
Wherein external shocks as instruments are examined and finite-sample bias under weak instruments is described, while two-stage estimation steps and diagnostic tests are outlined.
Mathematics without LaTeX
Wherein various methods for rendering TeX mathematics are surveyed, and client‑side libraries such as MathJax and KaTeX and server‑side pre‑rendering to SVG or HTML are contrasted.
Futurism
Useful ways of imaginatively forecasting
Wherein prediction markets and speculative designs by practitioners such as Anab Jain are examined as instruments for imagining societal trajectories, and statistical forecasting is briefly surveyed.
Learning Gamelan
Wherein a two-year project is described, in which convolutional networks are approximated by recurrent neural networks for time-series analysis, and poles of IIR filters are controlled by gradient descent.
Neural net attention mechanisms
On brilliance through selective ignorance
Wherein the structure of transformer stacks and self‑attention layers is described, the role in processing sequential data such as text is examined, and recent optimizations such as FlashAttention are noted.
ELBO
Evidence lower bound, variational free energy etc
Wherein the evidence lower bound is presented as a free‑energy decomposition and is shown to equal the expected log‑likelihood minus the KL to the prior, with ties to importance‑weighted sampling.
Privilege accountancy
Wherein the practice of mapping social advantage is chronicled and the origins of the privilege walk are traced to psychotherapy-infused critiques, situating power and marginalisation as a staged exercise.
Mechanism design for reputation systems
Karma, credit scores, pagerank, optimised ad hominem reasoning…
Wherein the design of reputational mechanisms is considered, attention being paid to iterative, PageRank-like ranking algorithms and to emergent, large-scale social-credit experiments in China, and to how platform gamification is mapped to real-world incentives.
Spreadsheetalikes
Wherein spreadsheet alternatives are surveyed and a 1700 BCE tabular artifact is invoked to situate the form, while web collaboration, real‑time streams, and R‑backed GUIs are noted.
Javascript mathematics
Wherein browser mathematics libraries are surveyed, probabilistic calculators and matrix toolkits are catalogued, and a shift toward WebAssembly and Python-in-browser via Pyodide is noted as shaping performance-oriented workflows.
Virtual private mesh networks
Pretending your phone is on your LAN
Wherein secure private meshes are surveyed for use in secure access rather than anonymity, and it is noted that tunnel brokering is performed by providers who observe which devices interconnect, authentication being routed via corporate identity providers.
Gaussian process regression software
Wherein implementations of Gaussian process regression are surveyed, and the availability of GPU acceleration, sparse variational inference, ecosystem bindings for Python, Julia, JAX and PyTorch, and examples of scaling to large datasets are noted.
Scenius
Wherein autopoietic outlier teams are examined as generators of collective creativity, the coinage is traced to Brian Eno, and the economist’s notion of efflorescence is adduced as a parallel.
Diversity in teams
Multiculturalism, pluralism and eccentricity at small scale
Wherein the economic arguments for team diversity are examined and the limits of productized diversity training, including evidence of backfire, the role of neurodiversity is considered, and data-driven interventions are critiqued.
Bayes linear regression and basis-functions in Gaussian process regression
a.k.a Fixed Rank Kriging, weight space GPs
Wherein the Gaussian process is represented in weight space by finite basis functions, and stationary kernels are approximated by Monte Carlo random‑Fourier features sampled from the kernel’s spectral density.
Simulating Gaussian processes on a lattice
Wherein the simulation is reduced to producing a Gaussian vector on an equally spaced lattice by circulant embedding and fast Fourier diagonalisation, so that the Toeplitz covariance is sampled efficiently.
Player vs game
Wherein the distinction between blaming the player and altering the game’s rules is examined, rather than faulting individuals, and stadium crowd‑crush incidents are invoked as a concrete motive for mechanism redesign.
Integrated Nested Laplace Approximation
Wherein the method is described as yielding approximate Bayesian inference for latent Gaussian models by applying nested Laplace approximations and exploiting sparse Gaussian Markov random field computations.
Red queen social signal dynamics
Arms races in memetic selection on graphs is how I make my fashion choices
Wherein social signalling is examined as a Red Queen struggle, and shibboleths are described as mechanisms to repel outsiders, with premium‑mediocre consumer taste invoked as a concrete example.
Bayes for beginners
Wherein prior selection, practical workflow for MCMC inference including Rao-Blackwellization, and teaching resources such as McElreath’s text are set out in concise, noncomprehensive notes.
Partial differential equations
Wherein Green’s functions, basis and Laplacian methods, and Eulerian approaches are surveyed, and fluid dynamical examples such as Navier–Stokes and CFD applications are noted.
Trauma and resilience
Wherein early responses to a Bad Thing are surveyed, and the timing of one‑off debriefing, sleep‑deprivation studies, and beta‑blocker trials is reported as altering later PTSD risk, with debriefing discouraged by WHO.
Data centric AI
Wherein the study of learning is reframed to prioritize curated datasets, and methods such as data versioning, benchmarking, semi‑supervised augmentation, and summarization are advanced as essential instruments
Julia, the programming language
The hippest way to get your IEEE754 on. Hngh.
From level 0 (julia-curious) up to level 2 (how do you overload broadcasting?)
Fun tricks in non-convex optimisation
Wherein the role of initialization and symmetries in steering final optima is examined, and phase retrieval is treated while a double-toboggan illustration is employed to show dependence on basins of attraction.
Australia
Wherein the nation’s complexities are surveyed, and colonial founding myths and recognition of traditional owners are noted, with historical Waratah tiles presented as an illustrative detail.
Empirical mode decomposition
Multiplying your exposure to uncertainty principles
Wherein a signal is rendered into intrinsic mode functions by an iterative sifting process, and instantaneous frequency content is obtained via the Hilbert transform for analysis of nonlinear, non‑stationary records.
Performance indicators, measurement, analytics
Wherein the art of measuring organisational performance is surveyed, a practice is described whereby metrics are recorded so that analytical work is judged by how quickly decisions are made, and Goodhart’s law is acknowledged.
Intro to probability
Wherein an interactive tutorial on probability is compiled, with Bayesian worksheets and computational tools such as Squiggle and Guesstimate being provided to enable hands-on exploration of probabilistic reasoning.
Astroturf and artificial reefs
On the sometimes-fungibility of status and cash in the realm of cultural cachet
Wherein the parallel is drawn between artificial reefs and engineered urban hipness, and globalized gentrification is examined, and 526 urban agglomerations above one million people are noted.
Comfy GNOME shell
Sparing thoughts for the desktop whatsit favoured by the thoughtless
Wherein GNOME shell is presented as Ubuntu’s default desktop environment, its mastery is described as non‑obvious, a catalogue of Super‑key shortcuts is provided, and low‑effort theming, tiling, and file‑manager fixes are outlined.
Hydrology, applied
Rivers, aquifers and other wet things that can flood your house
Wherein the quirks of installing MODFLOW are recounted and a solution by building MODFLOW‑NWT with the community pymake tool is described, enabling Linux and macOS execution.
Inverse problems
Wherein the reconstruction of hidden causes from observed effects is treated by Bayesian inference and regularized least squares, with applications to X‑ray crystallography, MRI and photogrammetry.
Betting
Wherein the special case of all-or-nothing wagers is examined, the Kelly criterion is invoked as a sizing rule, and the mechanics of exchanges and quantity market‑making are delineated.
Invasive arguments
Inflammatory topics, toxplasmic incidents. Weeds in public discourse
Wherein invasive arguments are depicted as contagions that reproduce by outrage, are explained via the toxoplasma of rage, and are shown to choke common discourse like kudzu.
Stochastic signal sampling
Discrete sample representation of continuous stochastic processes
Wherein the recovery of random continuous signals from discrete samples is considered, and posterior probabilities over sample paths are assigned under models including non‑Gaussian Lévy‑driven processes.
Signal sampling
Discrete representation of continuous signals and converse
Wherein signal reconstruction is examined via Nyquist rates and Hilbert-space projections, and nonuniform, compressed, and stochastic sampling regimes are contrasted for reconstruction error
Tokenism or table stakes?
“The least you can do” is the minimum unviable product
Wherein a disputation is set out about whether minor symbolic changes, exemplified by a demanded office honorific such as Spongtastic, are dismissed as posturing or are treated as minimal civic cost, and costs are tallied.
Social justice games, colonial games
Collective action for mutually assured destruction
Wherein coalition strategies are dissected, and the colonial tactic of divide-and-conquer, exemplified by British methods in India, is examined as an instance of elite-capture dynamics.
Tasmania
Wherein the island’s contested past is presented, an asserted ten‑thousand‑year oral tradition and debated losses and retentions of indigenous technologies being outlined in measured detail.
Generic dependency managers
Wherein generic dependency managers are surveyed and Vagrant is shown to instantiate disposable virtual machines via vagrant up to reproduce development environments, with a note on Spack for HPC.
Workhacks
How to succeed in business (with or without trying)
Wherein methods for surviving workplace bureaucracy and teamwork are set out, including the use of brag documents and asking for advice instead of feedback to advance low-rank workers.
Political axes, political correlations
Dimensionality reduction for the Great Society
Wherein political belief is treated as a multiaxial psychometric map, and correlations between extremism and rhetorical stridency are examined, with practical axes such as technocracy and prestige invoked.
Pluralistic ignorance, silent majorities, spiral of silence, hidden tribes
We all believe that we all believe what we do not believe
Wherein the phenomenon is described and illustrated by surveys at Reed College and an 8,000-person Hidden Tribes study whose social-media sharing rates are noted.
Institutions
Stable orbits in human systems
Wherein the collective neurosocial operating system by which societies are run is examined, and the management of commons together with the legibility of social arrangements is surveyed.
HDF5
A data format I need to know about
Wherein HDF5 is described as a late‑20th‑century scientific format, its built‑in compression is noted to perform poorly on floating‑point arrays and virtual datasets are supported.
Markov Chain Monte Carlo methods
Wherein Hamiltonian dynamics are invoked to guide proposals in samplers, tempering is described as a means to traverse modes, and coupling is employed to obtain unbiased estimators for parallelisation.
Markov decision problems
Wherein discrete-time stochastic control problems are presented, and partially observable cases (POMDPs) are examined, while connections to optimal control and learning of forward propagators are outlined.
Plants
Wherein the reader is informed that a Namibian plant called Welwitschia is reported to persist for a thousand years while bearing only two leaves, and botanical kinships are depicted as porous.
The Illawarra
The cheaper bit underneath Sydney comprising Wollongong etc
Wherein the coastal Illawarra is presented as Sydney’s overshadowed sibling, and the industrial harbour of Port Kembla together with the cascades at Bourke Falls are briefly noted for their local significance.
Machine learning and statistics in Julia
Wherein the Julia ecosystem is surveyed, Flux-based differentiable learning and DiffEqFlux-enabled Neural ODEs are noted, DataFrames.jl is cited as the standard for tabular statistics, and DSP.jl is mentioned.
Emergent spacetime
Wherein analogies to graph neural networks are noted, popular essays are surveyed, and quantum particles are invoked as candidate constituents from which spacetime is proposed to emerge.
Stochastic partial differential equations
SDEs taking values in some function space
Wherein stochastic partial differential equations are presented as Banach-space-valued stochastic differential equations driven by Q‑Wiener noise in multidimensional domains, and classic references are collected and imagery is included.
Applied psephology
Wherein the mechanics of voter modeling are examined, with Australian polling failures and ecological‑inference issues such as Simpson’s paradoxes in electoral demographics being considered.
Public speech norms as compatibility problem
Postel vs postal, legibility vs intelligibility
Wherein speech is treated as a technical compatibility problem, likened to USB connectors and network stacks, and the time costs of mastering dialectal registers and imposed tests are examined.
Funerals and other end-of-life stuff
Services, enduring guardianships, wills, burials cremations, esp in New South Wales
Wherein embalming in Australia is noted as uncommon and a solitary Southern Cryonics project is reported, while natural burial schemes and Muslim non‑coffin rites are outlined.
GUIs for numerical array data
Wherein HDF5 viewers are surveyed, and practical installation and memory‑use quirks are noted, including a default open action that attempts to load entire files into RAM.
Database and data file GUIs
Wherein a catalogue of database and data file GUIs is presented, and CastleDB’s storage of datasets as newline-separated JSON files enabling VCS-friendly diff and merge is recorded.
Crowd-sourced science
Wherein the globe is mapped by humble smartphones and forum posts, data being pooled through open-source field tools and online threads, and observations are processed by distributed volunteers.
Memetics
Taste dynamics, opinion dynamics, sincerely-held-belief dynamics etc
Wherein the propagation of belief is treated as an epidemiological diffusion, and Bass-style models are invoked to show how social selection and network structure determine which beliefs persist and which fade.
Doing email better, or better, not doing email at all
Wherein the perils and remedies of electronic correspondence are surveyed, and the practicality of self‑hosting SMTP servers such as postfix is considered alongside public‑key encryption’s usability limits.
System identification using particle filters
A.k.a. parameter estimation in data assimilation
Wherein the parameter vector is included in the state and is assigned a small random-walk evolution, particle filters are employed for joint state–parameter inference, and the evolution magnitude is left unspecified.
(Discrete-measure)-valued stochastic processes
Wherein stochastic processes on discrete measures are presented, and a construction is given whereby stationary Beta marginals are obtained from thinned autoregressive Gamma components to model allele frequencies.
Forecasting with model averaging
Mixtures of experts and regression ensembles applied to time series forecasting
Wherein model-mixing techniques are presented for time-series prediction, and predictor-conditional posterior densities for the next step are examined, with connections to dependent-data theory being noted.
Measure-valued stochastic processes
Wherein measure-valued stochastic processes are examined, and a dependent Dirichlet process is described in which stick-breaking weights are given by transforms of a stochastic process to induce indexwise dependence.
Linux audio
Making sound by banging rocks together 44 thousand times per second
Wherein the low‑latency kernel is recommended for realtime use, yet is found to provoke sporadic reboots on some NVIDIA‑equipped laptops, and PipeWire is presented as a unifying alternative.
knitr/RMarkdown etc
Wherein the analysis code and narrative are kept in continuous sync, figures are rendered and cached automatically, and web-ready reports are produced from a single markdown-based source.
Farming and husbandry of black swans and dragon kings
Heavy tailed and Knightian uncertainties for fun and profit
Wherein portfolio theory for outliers is considered, and allocation rules that are biased toward variance rather than precise measurement are examined as means to cultivate rare, high‑impact events in finance and philanthropy
Causal graphical model reading group 2022
Wherein Chapter 3 of Brady Neal’s course is recounted, potential outcomes are contrasted with causal DAGs, d-separation is developed to identify adjustment sets, and graphical rules are illustrated with example diagrams.
Inference without KL divergence
Wherein alternative divergences to KL are examined, and a probability-functional descent using von Mises calculus is presented for distributional and likelihood-free Bayesian inference, with algorithmic links to SGD.
SLAM
Simultaneous Location and Mapping
Wherein the problem of scene reconstruction by a moving camera is examined, and neural implicit representations with differentiable rendering are combined to enable scalable SLAM via least-squares inference.
Vecchia factoring of GP likelihoods
Wherein the Vecchia factoring is undertaken as an approximation in which the Gaussian process precision matrix is replaced by one whose Cholesky factor is rendered sparse, and likelihood computations are thereby cheapened.
Synchrony between things, especially organisms
Entrainment, synchronisation, dancing together
Wherein brain and bodily rhythms are observed to lock phase during music, conversation, and dance, and heartbeats and breaths are reported to align across co-present individuals, as studies describe.
Groupthink and the wisdom of crowds
Wherein the role of diversity and surprisingly popular polling methods in averting consensus errors is examined, signalling dynamics are analysed, and COVID-19 public communication is invoked as a case study.
Hierarchical models
DAGs, multilevel models, random coefficient models, mixed effect models, structural equation models…
Wherein hierarchical systems are introduced and a directed graph of interacting random processes is described, and inference of parameters and conditional distributions is pursued when observations are noisy and some variables remain unobserved.
Model fairness
Wherein causal accounts of discrimination, fairness‑accuracy trade‑offs, and feedback effects in lending and criminality prediction are examined, the concentric manifolds claimed for face‑based criminality are noted and post hoc interpretation methods are outlined
Social media if you must
Harm minimisation for corporate social network users
Wherein practical steps for escaping corporate networks are enumerated, including using mobile browsers with adblockers, scheduled cold‑turkey breaks, exporting personal archives, and quarantined single‑site browsers
Markov bridge processes
Especially Lévy bridges, Doob h-transforms
Wherein bridge processes are defined by conditioning Markov paths on fixed endpoints, midpoint marginals for Lévy increment processes are derived, and tractability for Brownian, gamma and Poisson bridges is examined.
Email clients
Wherein the landscape of email clients is surveyed, Linux-specific options are catalogued separately, and the availability of OpenPGP/GPG integration via plugins is noted.
Tracking my website traffic
Optimising scarce attention into the shopping cart widget
Wherein a minimalist tracker is considered, Gauges is employed at USD6/month and a script is provided to export site popularity data for local analysis, and geolocation is stored as estimates rather than raw IPs.
Social norms
Wherein the mechanics of conformity are recounted, pluralistic ignorance and Schelling points are noted, and the scaling of norm enforcement by automated observation—making transparent acts like jaywalking especially policed—is examined.
Application firewalls
Spyware mitigation and bandwidth management
Wherein application traffic is governed per‑process on macOS and Linux, and userland tools such as Little Snitch and OpenSnitch are cited to mediate outbound connections and limit bandwidth usage.
This is a simulation
Can the automaton learn to play the game of life?
Wherein the possibility that our observable universe is described by ~10^122 qubits is considered, Bekenstein bounds and conspiracy-analogies are invoked, and limits to inference are examined.
Beta Processes
Wherein Hjort’s Beta process is examined as a non-decreasing random measure and its relation to Lévy subordinators is queried, with connection to nonparametric random factor models being noted.
Stationary Gamma processes
Wherein six stationary time series with Gamma(α,λ) marginals and autocorrelation ρ^{|s-t|} are presented, among them a thinned AR construction using Beta thinning and a Poisson change‑point model.
Particle Markov Chain Monte Carlo
Particle systems as MCMC proposals
Wherein particle filters are introduced into MCMC samplers, the bootstrap particle filter is shown to yield Monte Carlo likelihood estimates and PMMH and particle Gibbs constructions are set forth for change‑point use.
Typesetting algorithms in LaTeX
Wherein the formatting of pseudocode in LaTeX is treated, with algorithmicx and algpseudocode used inside algorithm floats (noend option), line-number referencing exemplified, and minted noted for highlighted real code.
The levels of simulacra
Wherein four tiers of meaning are delineated, as an assertion that there is a lion across the river is parsed from literal report to partisan advantage, and is applied to pandemic and political rhetoric.
Myths
Placeholder for notes on stories we tell ourselves in order to make them true. Narrative with an action plan, which is to become real, in the social construction sense, and…
Probabilistic neural nets
Inferring distributions in neural nets
Wherein the inference of densities in massively parameterized neural nets is examined, and mixture density networks alongside ensemble and approximate Bayesian techniques are presented as means to quantify predictive uncertainty.
Plotting for the web
Wherein web plotting for data is surveyed, and interactive browser-based dashboards and JavaScript libraries (Plotly, D3, Vega) are examined, with emphasis on client-side interactivity and SVG output.
Playing music on the computer
Wherein the multitude of desktop players is surveyed, and the problem of exhaustive file indexing—dangerous to musicians with vast sample libraries—is noted and alternatives that avoid it are examined.
Beta and Dirichlet distributions
Wherein the Beta is presented as a ratio of independent Gamma variates and the Dirichlet is exhibited as their normalized vector, parameters being tied to Gamma functions and total concentration.
Attention economy
Wherein a multiscale system of agents with finite computational resources and limited attention is examined, and eyeball time is depicted as a contested, liquidity-like resource fought over for control.
Gumbel (soft) max tricks
Concrete distribution, relaxed categorical etc
Wherein the reparameterisation of categorical draws by adding independent Gumbel noise and a softmax temperature for annealing gradients to enable gradient-based learning via relaxed one-hot samples is described.
Change points
Looking for regime changes in stochastic processes. a.k.a. Switching state space models
Wherein an auxiliary run-time variate is introduced and online Bayesian run-length algorithms are described, and piecewise wide-sense-stationary segments are modeled for Gaussian-process extensions and real-time detection.
Pólya-Gamma augmentation trick
Wherein the Pólya‑Gamma augmentation is presented, an auxiliary variable is introduced so that the Bayesian logistic regression likelihood is rendered conditionally Gaussian, and Gibbs sampling is thereby enabled.
The language game
Coevolution of words and meanings
Wherein a toy model of colour words is examined in communicative context, connections to categorical stochastics and evolutionary models are adduced, and the persistence of shared terms is considered.
Detecting stationarity in stochastic processes
Change-points, trends and transients
Wherein a survey of methods for detecting non‑stationarity in stochastic processes is presented, change‑point, nonparametric and spectral approaches being outlined and ties to stability under fixed input distributions being noted.
Partition-valued random variates
Wherein random divisions of objects into unlabeled subsets are exhibited, and their interpretation as duals to categorical assignments and as exchangeable constructs used in Bayesian nonparametric models is noted.
Random binary vectors
Wherein distributions over n-length binary vectors are examined, and the existence of 2^n possible outcomes is noted, with continuous Gumbel-softmax relaxations and piano-roll representations being described.
Measure-valued random variates
Including completely random measures and many generalizations
Wherein random measures are surveyed and constructions such as completely random measures, Dirichlet and Gamma processes, and subordinators are presented, and conservation of mass in representations is considered.
Cloud ML compute vendors
Wherein cloud ML compute vendors are surveyed, and the cost and GPU availability are contrasted, and provisioning peculiarities and documentation sparsity are recorded, with surplus-market options like Vast.ai and OVH noted.
Reservoir Computing
Wherein reservoir computers are presented as chaotic dynamical systems that are programmed to emulate random‑access memory, virtual machines, and logic gates, and are trained chiefly by fitting linear readouts rather than by gradient descent.
Simulating Gaussian processes
Wherein methods for drawing finite-dimensional realizations from a specified covariance operator are described, and Lanczos-based Krylov subspace tricks and lattice and Langevin expedients are outlined.
Life-adjusted quality years
Whether to come to our party, if you like parties
Wherein a decision about attending a Sydney party is weighed, the COVID risk is put in micromorts and is assumed mitigated by same‑day RAT negatives and triple vaccination, and timing is compared to future uncertainty
Multivariate Gamma distributions
Wherein correlated Gamma vectors are constructed by Beta thinning and by a Lévy-measure representation on the unit sphere using parameters α and λ, and pairwise correlations are given in closed form.
Wiener-Khintchine representations
Spectral representations of stochastic processes
Wherein the covariance of weakly stationary processes is shown to be representable by a finite positive spectral measure on R^d, and the kernel is exhibited as the Fourier dual of the power spectral density.
Sparse coding
Wavelets, matching pursuit, overcomplete dictionaries…
Wherein the representation of signals is sought by linear expansion in learned or redundant dictionaries, the noisy-observation case being emphasized and algorithms such as matching pursuit being discussed.
Survival analysis and reliability
Hazard rates, proportional hazard regression, life testing, mean time to failure
Wherein the peculiarities of right‑censoring are examined and the hazard and cumulative hazard functions are defined and connected to survival probabilities for estimating lifetimes.
Plotting in R
Wherein R’s plotting capabilities are surveyed, and it is noted that plots are served via Shiny to power web applications, ggplot2 is emphasized with extensions and interactive editors such as ggedit
Self-supervised learning
I just wanna be meeeeee / with high probabilityyy ♬♪
Wherein a notebook on self-supervised learning is presented, with emphasis placed on contrastive learning methods, and illustrative notes and figures on transformed signals are supplied.
Race, politics of
Wherein the tangled claims about race are surveyed, and the uneasy interplay between genetics, statistical classification, and political discourse is set out, including a taxi driver’s conflation of religion and race and non‑US contexts.
Opinion dynamics 1: Social contagion moves hearts and minds
How to win elections and influence people
Wherein political opinions are treated as contagious through interpersonal networks, an analogy to John Snow’s cholera map is employed, and the notion of social penumbra is invoked as a transmission mechanism.
Fun with rotational symmetries
Wherein methods for radial functions are explicated, integrals on n‑spheres and n‑balls are reduced to univariate forms, Hankel‑type transforms and random sampling procedures are exhibited.
Matrix- and vector-valued generalizations of Gamma processes
Wherein matrix and vector Gamma processes are considered, and a tractable AΓ family is exhibited, for which explicit formulas for mean and covariance are provided in terms of Σ, η and ω.
Lévy Gamma processes
Wherein the Lévy Gamma process is presented as a subordinator with independent Gamma increments, its Lévy measure π(x)=α e^{-λ x}/x is exhibited, and the Gamma bridge via Beta thinning is described.
Gaussian processes on lattices
\[…
Institutions for devils
Wherein institutions for devils are examined and the role of economic mechanism design in anticipating selfish sociopaths who will attempt to change the game is considered, and design failures are noted.
OODA loops
Wherein the Observe–Orient–Decide–Act loop is examined as a rapid decision cycle and is applied to map-and-compass navigation training, while its timely sensor interpretation and feedback dynamics are outlined.
Subordinators
Non-decreasing Lévy processes with weird branding
Wherein a non-decreasing Lévy process is described as a subordinator, its Laplace exponent and Lévy–Khintchine decomposition with drift and jump measure are evoked, and its role as a random time change is noted.
Stability in dynamical systems
Wherein the parameterization of systems is examined and Lyapunov exponents are invoked to detect sub-superpolynomial growth, and linear cases are reduced to polynomial root problems.
Adversarial learning
Wherein the noise is construed as worst-case within given constraints, contrasted with random perturbations, and is related to game-theoretic tactics and Goodhartian target gaming.
Audio sample management
Wherein audio samples are catalogued by machine-listening algorithms, and 2D similarity maps are employed to surface drum hits and foster serendipitous discovery within production workflows.
Moral calculus
Wherein trolley problems and machine agency are examined, and weaponised 3D‑printable golems, autopilot ethics, and continuous branching decision‑tree limits are invoked to probe moral choice.
M-estimation
Wherein M-estimation is presented as estimation by extremizing loss functions, its ties to loss‑based machine learning and large‑sample asymptotics are noted, and influence functions are sketched.
Our eating disorder
Wherein Western fad diets, bourgeois ethics of eating, marketing and body‑image are catalogued, and the recent turn to GLP‑1 diabetes drugs for weight loss is presented as a practical angle.
Politics as statistical learner
Wherein politics is treated as a statistical learner, and the centrifugal governor is invoked to show how institutional feedback and loss functions are calibrated across censuses and public experiments.
Neurons
Neural networks made of real neurons, in functioning brains
Wherein neural computation is presented at an intermediate scale and is described as being driven by discrete spikes in continuous time, with heterogeneous cell types and messy organisation being noted.
Myopic optima in morality
Blowbacks, reverse psychology, norm enforcement
Wherein examples of naïve interventions are examined and perverse outcomes are exhibited, as in teen pregnancy, trauma-amplifying policing, offensive-speech rebound, drug prohibition, and allergy-avoidance backfire.
Ergodicity and mixing
Things that probably happen eventually on average
Wherein ergodicity and mixing are examined, and mixing conditions such as β‑ and ϕ‑mixing are related to finite‑sample learning guarantees for dependent data, and to Lyapunov exponents measuring sensitivity.
Window management in macOS
Wherein macOS window arrangement is surveyed: Mission Control shortcuts are reported, side-by-side Split View is noted, and third-party tiling tools from Divvy and Amethyst to yabai and scriptable Hammerspoon are catalogued.
Tiling window managers
Desktop management how Pajitnov intended
Wherein the screen is partitioned into non‑overlapping panes and keyboard‑driven workflows are explored, and the tension between X11 and Wayland support for such managers is surveyed.
Diversity in society
Pluralism, multiculturalism, tolerance, ghettoisation, xenophobic panic
Noisy chaos of notes about the frictions between people in the presence of differences in culture, ethnicity, sex, sexuality, neurotype, etc… Notes on how the flames of…
Visual node based programming
a.k.a. dataflow graphs, patchers, visual coding, flow-based programming
Wherein node‑based visual programming is presented as a graphical dataflow system whose elements are mapped to textual code, and whose applications are exemplified by audio synthesis and neural‑net workflows.
Experimental ethics and observational data
Wherein challenge trials, ethics approvals, and observational surveillance are examined, and it is argued that experimental ethics combined with observational data render a pervasive surveillance state thereby demanded.
Variational inference
On fitting something not too far from a pretty good model that is not too hard
Wherein stochastic gradient descent, Monte Carlo gradients, and message‑passing are met with variational inference, and amortization is observed to produce variational autoencoders as a practical instantiation.
Probabilistic graphical models
Wherein factor graphs, plate notation and hierarchical, per‑group latent variables are surveyed, and the role of plates in representing thousands of local and global parameters for scalable inference is explained.
Random graphical models
Causality in amongst confusion
Wherein priors over causal graphs and sparse interaction structures are considered, and a catalogue of random models — from random neural feature maps to trophic and sparse hypergraph ensembles — is presented.
Evolution
Wherein the mechanics of heredity are surveyed, stochastic models of allele diffusion and replicator dynamics are examined, neutral and quasispecies models are noted, and links to optimisation are delineated.
Cryptocurrencies
Imagine if keeping your car idling 24/7 produced solved Sudokus you could trade for heroin
Wherein the practical role of cryptocurrencies in low‑friction cross‑border transfers is examined, an 8% effective fee on an Australian purchase is recorded, and the difficulty of using them as tips is noted.
Biological phylogeny
Wherein the history of life is presented as a branching diagram, and the Coelacanth is noted to be a 190 millionth cousin, 100 million times removed, while the tree metaphor is subjected to scrutiny.
Teaching
Wherein the practice of instruction is considered, and the efficacy of active learning versus traditional lecturing is reported, student perceptions are contrasted with test outcomes, and resources for undergraduate pedagogy are catalogued.
Outsourcing, applied
Wherein various forms of outsourcing are surveyed and practical venues for hiring researchers and assistants are catalogued, including Upwork, Double, TaskRabbit, and tools such as Hubstaff for time‑tracking.
Sociology and politics of information
Epistemic democracy, cognitive democracy, the great society
Wherein collective learning and contagion models on networks are examined, and the susceptibility of institutions to memetic spread and misinformation is delineated through Bayesian and evolutionary frameworks
Neural nets with basis decomposition layers
Wherein neural nets with basis decomposition layers are examined, and continuous analytic bases are proposed to enable native interpolation, autodifferentiable spatial gradients, and application to learning partial differential equations
Here’s how I would do art with machine learning if I had to
Wherein plausible deniability is afforded to the author for producing generative works by running trained neural models in reverse, notably via CLIP+GAN or diffusion pipelines, whilst attending to mathematics.
Karhunen-Loève expansions
Wherein a stochastic process is expressed via the covariance operator’s orthonormal eigenfunctions, and ordered eigenvalues are used to scale coefficients so that the Karhunen–Loève series is obtained.
Running neural nets backwards
Wherein methods for treating neural networks as inverse maps are examined, and applications to ill‑posed inverse problems and regularization techniques such as DeepDream and reversible architectures are delineated.
Spatial processes and statistics thereof
Wherein spatial processes are considered over continuous supports, and Gaussian process regression, termed kriging, is shown to be made scalable by fixed‑rank basis decompositions and SPDE‑GMRF approximations.
Bootstrap
Shuffling reality to produce your data
Wherein resampling of one’s data is employed to estimate an estimator’s sampling distribution and to correct bias, and where adaptations for dependent time series and for Bayesian variants are noted
Learning on manifolds
Finding the lowest bit of a krazy straw, from the inside
Wherein learning on prescribed curved spaces is considered, stochastic processes are invoked, and optimisation on the manifold of positive‑definite matrices is treated.
Quantum computing
Wherein the pursuit of quantum computing is narrated and the recent emphasis on quantum supremacy and software stacks such as TensorFlow Quantum for ML-enabled simulation is reported.
Feedback system identification, linear
Wherein feedback system identification is addressed for irregularly sampled data, and continuous‑time autoregressive models are fitted via Kalman recursion and transformed‑coefficient methods, while likelihoods are observed to be multimodal.
(Outlier) robust statistics
Wherein outlier-robust statistics are surveyed and M-estimation with Huber loss is presented, while corruption models — ε-contamination, adversarial total-variation, and Wasserstein perturbations — are delineated.
E-readers
Very expensive paper substitute that breaks if I drop it
Wherein e-readers are described as e‑paper tablets whose long‑lasting batteries permit extended off‑grid use, and whose annotations and files are synchronized to desktop libraries via syncthing to Zotero, an Onyx Boox Note Air 2 being cited as exemplar.
Dunning-Kruger theory of institutions
Lay theories of social mechanisms
Wherein institutional failures are cataloged and the misapplication of simple models is exposed, with system‑dynamics simulations such as the Beer Game presented as illustrative evidence and policy learning reframed as statistical learning.
Dunning-Kruger theory of society
Wherein the Dunning-Kruger effects on collective judgment are examined and experimental evidence is cited showing popularity, not quality, is often amplified by social transmission.
Care and feeding of macOS filesystems
Wherein the maintenance of macOS filesystems is treated as a practicum, and instructions for TRIM enabling on third‑party SSDs, rsync iconv quirks, and command‑line trash utilities are provided.
Docker containerized apps (for scientists)
Doing things that previously took 0.5 computers using 0.4 computers
Wherein Docker for scientific workflows is laid out, the Dockerfile recipe is described, GPU support challenges are noted, and an opaque registry timeout with a Google DNS workaround is reported.
Software package managers
Wherein the manners of installing necessities are surveyed, and the Nix method is described, whereby packages are built without side effects and are recorded as uniquely hashed entries in /nix/store.
Risk perception and communication
Wherein the quantification of hazards by units such as micromorts is presented, fat‑tailed and exponential dangers are examined, heuristics that skew public response are surveyed, and communication methods are outlined.
Sex and sexology
Incorporating smut, lewdness, and prurience
Wherein archival sexology collections and Victorian anti‑masturbation tracts are surveyed, and modern search‑data and dating‑site archives are invoked to trace changing sexual discourse.
Faust, the DSP language
Wherein the language’s compiler is described as translating DSP specifications into C, C++, WebAssembly, Rust and LLVM, and optional IDE and Python bindings are noted, while buffer access via the table keyword is highlighted.
Bounded rationality
Plus miscellaneous rationality postulates restricting von Neumann-Morgenstern
Wherein prospect theory and heuristic models are surveyed, computational limits of choice are invoked, and applications to neuromarketing, market design, and group decision are noted.
Social psychology
Which of those NPR-friendly studies actually replicated?
Wherein the limits of social psychology are surveyed and the replication crisis and priming effects are scrutinized, while the rise of data‑mining elites is noted as reshaping societal models.
Bayesian inverse problems
Wherein a hierarchical Bayesian formulation with unknown regression parameters is presented, and joint inversion for latent inputs is treated via posterior learning from training pairs and Laplace approximations for functional, high‑dimensional problems.
Gamma-Beta algebra
Wherein the Mellin transform is used to represent laws via quotients of Gamma-function products as ratios of Beta and Gamma variables, and parameter cancellation plus multiplicative composition rules are exhibited.
Categorical random variates
Wherein categorical random variates are surveyed and techniques such as stick‑breaking constructions, Gumbel‑max perturbations for argmax sampling, and Dirichlet‑process priors with Pólya‑Gamma augmentation are described.
Hardened desktop operating systems
Also amnesiac and/or anonymous
Wherein open-source desktop systems are surveyed and an amnesiac live‑USB and VM‑based Tor routing approach is presented as a tactic for journalists in hostile states, while distrust of hardware and build chains is noted.
Statistical projectivity
Wherein the notion of projectivity in statistical models is examined as a sometimes implicit property, and is linked to conditional consistency of finite-sample distributions across varying sample sizes.
Standards hell
Lock-in, QWERTY, fragmentation problems etc
Wherein the propagation of technical standards is examined through Postel’s robustness principle and path‑dependence, and the tension between federation and centralisation is illustrated by messaging protocols and power‑outlet examples.
Models of the mind
Wherein the mind is treated as a machine-learning engine and predictive coding is examined, accounts of how an algorithm feels are given, and depressive states are presented as failures of inference.
Presentations
Slide decks and other stylised academic dominance displays
Wherein methods of giving PowerPoint talks are outlined, harm minimisation is advocated, assertion–evidence aesthetics are invoked, recordings are relied upon for slide utility, and pitching to lay audiences is considered.
R packaging, installation etc
Wherein a method for installing R on macOS via Homebrew cask is prescribed, Apple Silicon makevars adjustments are given, and renv is recommended for project-local dependency management.
Matrix-valued random variates
Wherein matrix-valued random variates are surveyed, and distributions for positive-definite covariance matrices, notably LKJ priors for correlation matrices via Cholesky factors, and random rotations are presented.
Stickers
Printouts that you do not lose because they are adhered to a very large thing
Wherein hexagonal promotional stickers are described, tessellation and standardized dimensions are noted, and incompatibility between some print-on-demand and hexbin specifications is remarked.
Music software frameworks
and programming languages, for music
Wherein rapid prototyping of audio algorithms is favoured, production deployment is distinguished from installation use, and tools such as the SuperCollider scsynth backend and JUCE plugin host are noted.
Apple laptops
Wherein the desire for macOS‑only audio plugins is described, and the procurement of a refurbished or second‑hand MacBook with upgradeable storage or Apple Silicon compatibility is examined.
Backups
Version control for horrible data
Wherein encrypted backup solutions are surveyed, and a preference for tools such as Restic is indicated, with the alternative of Tarsnap noted as an offsite service costing roughly $0.25 per GB per month.
Group size
Scaling dynamics and network effects in social coordination
Wherein group size is treated as being constrained by cognitive limits such as Dunbar suggested neocortex-derived thresholds, and trade-offs with legible governance structures are sketched.
Top influences of 2021
Content that changed my life this year, and which also might change yours
Wherein a year’s influential writings are collected, and their effects on the author’s judgment and practices, notably Cassie Kozyrkov’s decision theory and the microCOVID risk tool, are quietly recorded for future reckoning.
Time frequency analysis
Multiplying your exposure to uncertainty principles
Wherein the Bayesian provision of probabilistic spectral analysis is considered, and locally stationary windows are assigned distributions for adaptive time–frequency decomposition.
Garbled highlights from NeurIPS 2021
Wherein workshops on machine learning for the physical sciences are catalogued, an online variational filtering method and a Laplace PyTorch library are reported, and a NeurIPS paper-visualization is noted.
Diversity as an end in itself
On a rich and vibrant ecosystem of culture and thought
Wherein it is considered whether cultivating human oddity is to be pursued as an end in itself, and the probing of a configuration space of possible intelligences is proposed as a constituent aim.
Causality via potential outcomes
Neyman-Rubin, counterfactuals, conditional treatment effects, and related tricks
Wherein the potential-outcomes approach is presented as a practised statisticians' method, its ties to Neyman and to Pearl’s DAGs via Single World Intervention Graphs are outlined, and disputes of lineage are noted.
Gradient descent, Newton-like, stochastic
Wherein stochastic Newton-like updates are described, and subsampled or unbiased Hessian estimators are employed to compute inverse-Hessian-vector steps via Hessian–vector products for large-data training.
Variational state filtering
Wherein a global, telescoping variational approximation is presented, by which latent states and system parameters are jointly estimated and per-step bias in sequential filtering is reduced.
Bundled/ packaged apps for Linux
Wherein the merits of one‑build, cross‑desktop distribution are chronicled, and the practical tensions between sandboxed permission models and modular plugin deployment are observed, and update and disk‑space behaviors of Snap and Flatpak are compared.
Convolutional subordinator processes
Wherein stochastic processes are defined by convolution of Lévy subordinators with smoothing kernels, and nonparametric distributions over measures are thereby produced by the kernelized noise.
Random rotations
Wherein uniform and tiny random rotations are considered, Haar measure is invoked and second and fourth moments of matrix entries are computed, and Givens rotations are presented as block perturbations of the identity.
Strategic ignorance
Wherein a practiced restraint of knowledge by researchers is described, and the deliberate omission of data is employed to avoid bias in inquiry, and subtle effects on collaboration and interpretation are examined.
Multi-output Gaussian process regression
Wherein a unifying view of vector Gaussian processes is presented, and a low-rank linear mixing of scalar GPs is assumed to enable scalable multi-output inference for time-series, with exact inference methods and toolkit pointers provided
Sydney food suppliers
Practical bulk budget Sydney gourmet
Wherein the means of provisioning in Sydney are delineated, and bulk purchasing options for a seven-person household and a cheese-maker serving warm buffalo ricotta are disclosed.
Gaussian Processes as stochastic differential equations
Wherein Gaussian processes are recast as stochastic differential equations via spectral factorization into finite-dimensional state-space models, enabling Kalman-style linear-time filtering for Matérn and rational kernels.
Ensembling neural nets
Monte Carlo with pre-rolled dice
Wherein the practice of ensembling neural networks is recounted, with dropout presented as an implicit ensemble approximating a deep Gaussian process, and BatchEnsemble tricks are adopted due to GPU constraints.
t-processes, t-distributions
Wherein t-process priors for regression are examined, and it is shown that conditional degrees of freedom increase with sample size, so that posterior behavior is rapidly approximated by Gaussian processes.
Convolutional neural networks
Wherein convolutional neural networks are presented as a topology whose layers are constructed with finite-impulse-response filters and pooling, receptive fields are computed for analysis, and activations are visualised as high-rank tensors with exploitable regularity.
Gamification
“Belated Blogging of a Buzzword” achievement unlocked
Wherein the human appetite for arbitrary goals and intermittent rewards, likened to sparse reinforcement‑learning signals, is examined as being commodified and applied to commuting, social media, and education.
Lévy processes
Stochastic processes with independent increments, jump diffusion
Wherein the class of continuous-time processes with stationary independent increments is considered, and a canonical decomposition into drift, Gaussian covariance and a Lévy jump measure is displayed.
Financing utopia
Wherein various proposals for funding capital‑intensive projects are examined, including worker‑owned firms, crowd‑lending instruments and tokenized micropayments as institutional experiments.
Cooperation amongst humans
Wherein the global blogging infrastructure is employed to examine mechanisms of human coordination, from antisocial punishment and status dynamics to pathogen-driven cultural boundaries and interacting learning modes
Energy-based models.
Wherein it is shown that energy-based models are treated as unnormalized probability functions, trained by contrasting model and data samples, and sampled from by MCMC-like dynamics to avoid normalization.
Digital nostalgia
Pixel art, geocities chic, cyberpunk retrofuturism
Wherein various forms of digital nostalgia are surveyed and the practice of running Doom on obsolete Kodak cameras is presented as emblematic of preemptive obsolescence and aesthetic repurposing.
Sexy plants
Wherein Linnaeus’s floral anatomy is described in domestic bed metaphors, the British scandal over sexualized plant classification is chronicled, and fungal multiplicity of sexes is noted
Teaching and doing mathematics remotely
Wherein remote mathematical instruction is treated and recommendations are given, including low-bandwidth recording strategies, endorsement of privacy-preserving Jitsi, and use of math-aware chat and virtual whiteboards.
Spectral graph theory
Wherein localized Chebyshev filters are defined via the graph Laplacian, are applied to network signal processing and graph neural networks, and eigenvalues of connectivity matrices are computed to inform filter design.
Better homes and societies
Wherein a scrapbook of links and aphorisms is presented, and reproducible science is reimagined as queer anarchism while activist etiquette and movement design are considered for governance and solidarity.
Probably Approximately Correct
Wherein a theory of learning is examined and PAC‑Bayes risk bounds are presented, emphasizing aggregated and randomized predictors and recent applications to neural networks by Dziugaite and Roy
Survey modelling
Adjusting for the Lizardman constant
Wherein survey practices are examined and the limits of inference are delineated, with attention to design, post‑stratification adjustments and a noted irreducible ~4% noise floor in responses
First aid
If you are checking this webpage while your colleague bleeds out, you are doing it wrong
Wherein common measures for bodily emergencies are set forth, attention being given to psychological first aid taught in COVID courses and to distinctions between household superglue and medical cyanoacrylates.
Eyewear
Wherein a pair of Emgo Bandito goggles is procured and the Dresden Vision system with interchangeable frames and a ten‑year warranty is described, and the consolidation of brands under Luxottica Essilor is noted.
German
Wherein notes on learning German are presented, listening resources and etymological tips are catalogued, SWR2 radio plays are recommended, and an interest in Yiddish influence is noted.
Networks and graphs, theory thereof
Wherein the relationship between electrical network conductance and random walks on graphs is examined, and graphons are presented as continuum limits connecting discrete ensembles to analytic functions
Economics on networks
Wherein models are presented in which economic interactions are constrained by graph ties, and methods for inference on social graphs are surveyed, with pointers to notable researchers.
Neural music synthesis
Wherein neural music synthesis is surveyed and differentiable DSP methods and raw time‑domain models are noted, with pointers to diffusion, Jukebox, and waveform‑domain approaches.
Fun with determinants
Especially Jacobian determinants
Wherein standard determinant identities are recited, the determinant-as-product-of-eigenvalues and scaling laws are stated, expansions of det(I+ A) for n=2,3,4 and a small-ε trace‑based approximation are presented.
Random neural networks
Wherein untrained neural networks are treated as functional artifacts, and random recurrent reservoirs are presented as feature factories whose steady states are used to fit downstream classifiers without training.
Dual booting MS Windows and linux
Wherein dual booting is described as requiring maintenance of two OSes, timekeeping compromises are imposed, shared data is advised to reside on NTFS, Windows fast startup is to be disabled, and Bluetooth is found to require recurrent re‑pairing.
Printing
Offline backups that do not need batteries
Wherein simple printing in Sydney for non-designers is chronicled, Flash-based upload interfaces are required, courier postage is charged for some services, and Canva’s print option is used to avoid full-page bleed.
Planning under uncertainty
Transcending planning-though-hope-and-haruspicy
Wherein deliberations are presented on consulting expert probabilistic judgments and on adopting preemptive measures like quarantine when modest chances of catastrophe are inferred, and statistical models are treated as pragmatic tools.
Multilinear algebra
Outer products, tensors, einstein summation
Wherein tensorial concepts are treated with index notation and Einstein summation, and matrix differential techniques for deriving gradients are demonstrated to simplify multilinear computations.
Missing data
Imputation, estimation despite etc
Wherein the phenomenon of missing data is presented through an illustration of a vanished cat, and a bibliographic reminder to read Morvan 2021 is appended for later review.
Practicalities of regularising neural networks
Generalisation for street fighters
Wherein the practicalities of regularising neural networks are surveyed, and methods such as early stopping, stochastic weight averaging and weight/spectral normalization for ill‑conditioned RNNs are described.
Fractals and self-similarity
Wherein noninteger Hausdorff dimensions are examined, iterated function systems and fractional derivatives are treated, a compression-based estimator for fractal dimension is sketched, and links to long-memory processes are noted.
Arpeggiate by numbers
Workaday automatic composition and sequencing
Wherein arpeggiation is treated as numerical selection of MIDI pitches rather than timbral design, and geometric, neural, and tool-based approaches are catalogued.
Hygienic masks
Wherein the distinctions between hygienic and particulate masks are set out, with attention to differing priorities of fit, material and sanitation, the presence of release valves, and cited evidence regarding wearer protection.
Approximate Bayesian Computation
Posterior updates without likelihood
Wherein Bayesian computation is presented as being effected via simulation-based inference when the likelihood is unavailable, and sequential Monte Carlo and neural methods are noted as applied.
Heavy tails
Weird things about rare massive events
Wherein the peculiarities of distributions with power-law decay are recorded, and the tendency of extremes to dominate sample means, yielding divergent variances, is exemplified.
GIFs
Wherein the reader is apprised that animated GIFs are handled via FFMPEG commands, and that a typical workflow is shown—for example extracting seconds 5–8 and producing a 10fps, 480‑pixel GIF.
Software engineering for scientists
Wherein reproducible practices are surveyed, and the emergence of executable papers and toolchains such as CodaLab and PapersWithCode is chronicled to enable papers to be paired with runnable code and datasets.
Meta learning
Few-shot learning, learning fast weights, learning to learn
Wherein meta-learning is presented as methods for few-shot adaptation, the 1990s idea of neural nets programming neural nets via fast weights is noted, and the utility of inner-loop updates is questioned.
Biological basis of language
Neurology, evolution and ecology of our memes
Wherein the neural foundations of language are surveyed in a laconic register, and the study of finch syntax is adduced as a concrete model, while links to predictive coding and transformer analogies are outlined.
Natural language processing
Automatic processing of words and sentences and such
Wherein the computational study of human language is treated, and attention‑based neural architectures for translation, parsing, generation, and semantic inquiry are described.
Fractional differential equations
Wherein fractional derivatives are introduced via Laplace-transform representations, and non‑Markovian memory effects are thus incorporated into differential equations, with applications noted in pharmacokinetics.
User interface design
Wherein the ergonomics of control layouts and the user experience of human interaction are surveyed, and a notorious volume-control layout and an interactive anti‑pattern experiment are cited.
Maths hacks
Wherein miscellaneous metamathematics is surveyed and practical problem-solving tricks are compiled, exemplified by matrix differentiation and the Cauchy residue trick for spectral analysis, with learning and pedagogy links.
Software video routers
Wherein virtual camera bridges and GPU-backed frame sharing between applications are described, and tools for creating virtual webcam inputs are catalogued, including OBS and Syphon and Spout pipelines.
Wirtinger calculus
It’s not complicated / It’s complex
Wherein Wirtinger calculus is presented as a method for differentiating real-valued functions of complex arguments, and is shown to be employed in optimisation tasks in signal processing, notably phase retrieval.
Recurrent neural networks
Wherein recurrent neural networks are described as feedback systems with a hidden state and memory, their use in signal‑processing and links to linear systems, LSTM gating, reservoir computing, and attention are sketched.
Governance of the commons
On the mysterious fact that most people get on with most of their neighbours most of the time
Wherein the governance of shared resources is examined through Ostrom’s institutional frameworks, a taxonomy of public goods is outlined, and online discourse is treated as a commons to be managed.
Path smoothness properties of stochastic processes
Continuity, differentiability and other smoothness properties
Wherein conditions for Hölder modifications are set out via Kolmogorov’s moment bounds, and Gaussian sample paths are shown to lie outside their RKHS with probability one when the RKHS is infinite dimensional.
Stochastic calculus
Calculus that works, in a certain sense, for random objects, of certain types. Instrumental in stochastic differential equations. This is a popular and well-explored tool…
VS Code as R IDE
Wherein VS Code is outfitted as an R IDE and an in-editor R session is enabled via vscode-R, httpgd is employed for browser-based plotting, and languageserver is required for deep integration.
Baby’s first private cloud
Wherein a homebound private cloud for hobbyists is described, containerized apps and VMs are kept alive by simple orchestration (Docker Swarm, Nomad) and external access is mediated by VPN or Tailscale.
X11, Wayland etc
The other antiquated windowing system
Wherein window manipulation and simulated input are examined, it is noted that xdotool operates via X11’s XTEST extension to send keystrokes and mouse events, and that ydotool exists for Wayland.
Vector Gaussian processses
Wherein vector Gaussian processes are introduced via matrix-valued cross-covariance functions and two notions of positive-definiteness, and an extended input space is proposed to handle multiple outputs.
Convolutional stochastic processes
Wherein convolutional stochastic processes are described as moving averages of white noise, exemplified by Gaussian and subordinator convolutions, and their relation to kernel smoothing is noted.
Essays in stochastic processes
My PhD thesis with Zdravko I. Botev
Wherein a doctoral thesis in stochastic processes is presented, audio style‑transfer experiments with trumpet spectrograms are included, and the full thesis is made available for download.
Virtual private networks
More options for internet privacy
Wherein the uses of virtual private networks are surveyed and the Linux DNS‑leak quirk of OpenVPN is noted, and choices between DIY servers, commercial providers, WireGuard, auto‑reconnect and local‑network bypass fiddliness are described.
Design grammars
Wherein L‑systems are described as grammars for generating plants, seashells, music and dungeons, and inverse procedural modelling is treated as their inference counterpart, with links to fractal image compression.
Grammar induction
Wherein the task of inferring formal languages is treated as the recovery of syntactic rules with probabilistic measures, and the extension to design grammars for nonlinguistic artifacts is considered.
Risk neutral measure
Wherein a probability is selected so that discounted asset prices are rendered martingales under the chosen numeraire, and option payoffs are then valued by expectation taken under that measure
Hardened mobile
Trusting the computer that follows you around all day
Wherein the mobile twin to hardened desktops is examined; methods for reducing phone spyware via accountability are outlined, hardware options like Librem 5 and Precursor are surveyed, and on‑device CSAM scanning is noted.
Generalized linear models
Wherein the familiar least‑squares apparatus is extended to handle non‑Gaussian responses through link functions and quasi‑likelihood, and additive, hierarchical and vector extensions are outlined.
GPU computation
Wherein options for SIMD on GPUs are surveyed, including numba and RAPIDS, and the practicality of renting pre‑configured cloud GPUs rather than maintaining local machines is noted.
Sequential experiments
Especially multiple sequential experiments
Wherein the conduct of successive trials is described, and strategies for minimizing time wasted on negative outcomes are pursued, while Bayesian optimization is invoked to identify superior interventions
System76 laptops
Wherein a compact System76 Lemur is described, supplied with a comprehensive user repair manual and Pop!_OS configuration, and four-hour battery endurance under a demanding development workload is noted.
Unix/linux distros explained as bikes
Wherein the Linux distribution landscape is mapped to bicycles, and each distro is described by concrete bike quirks: Ubuntu is delivered magenta, Debian is shipped without wheels, and NixOS is presented with reconfiguring drivetrains.
Games, computer, recreational
Wherein a 2D platformer with integrated scripting and level designer is recorded, retro titles are observed to be runnable in browsers, and bespoke arcade hardware is noted as difficult to emulate.
Contact tracing
Wherein the mechanics of proximity versus location methods are considered, and privacy-preserving, Bluetooth-based tracing is described as being weighed against coarse, identifying GPS approaches.
Generic cloud machines
Wherein cheap ARM RHEL instances and frugal block storage for ephemeral virtual machines are examined as alternatives to large-provider orchestration, OVH and Oracle Free Tier being noted.
IDEs for R
Friendly UIs for the almost-friendly statistical programming language
Wherein various interfaces for R are catalogued and compared, and it is noted that RStudio can be served as a remote web app with an integrated graphical debugger, alongside consoles like radian and Jamovi.
Contagion processes and their statistics
Wherein contagion between georegions and networked populations is surveyed, multivariate point processes such as Dirichlet‑Hawkes are invoked, and the problem of identifiability under noisy, incomplete observations is outlined.
Media virality
Strategic modelling for content creators
Wherein the role of short-form platforms such as TikTok in rapid idea contagion is examined, and archival references from the ANU Computational Media Group are catalogued for further excavation.
Learning summary statistics
Wherein dataset-level summary statistics for likelihood-free inference are considered, and their joint tractability with simulation-to‑observation distance measures is examined, with neural‑statistician and deep‑sets constructions noted.
GPU computation out of the cloud
How is deep learning awful this time?
Yak shaving risk.
Multi-task ML
Wherein multi-task models are presented as means to produce multivariate predictions from univariate losses, and Gaussian process formulations are described as a natural multivariate extension.
Tensorflow
The framework to use for deep learning if you groupthink like Google
Wherein TensorFlow is described as a C++/Python neural-network toolkit by Google, and its ecosystem is recounted as convoluted with legacy APIs and installation friction, though Edge ML tooling is noted as a surviving practical advantage.
Point process intensities and statistical estimation thereof
Wherein the intensity of inhomogeneous Poisson point processes is examined and kernel and likelihood-based estimation techniques are presented for recovering spatially varying event rates.
Graph sampling
Estimating functionals of graphs
Wherein various sampling procedures for social networks are surveyed, and edge-exchangeable and paintbox processes are invoked to characterize inference limitations under projective and design-unbiased schemes.
Semi/weakly-supervised learning
On extracting nutrition from bullshit
Wherein methods are surveyed for learning true labels and annotator reliability from noisy crowd labels, using hierarchical generative models or graph-based label propagation and data augmentation tools.
Applied string mangling
Regexes, parsing, tokenising etc
Wherein common string-mangling strategies are surveyed; a duplicate-word regex is presented and tools for code-aware rewriting such as Comby and parser generator SLY are catalogued.
Moral philosophy
Wherein the tension between spontaneous heroism and earlier civic choices, as when a rescuer had voted against levee reinforcement, is examined and biological constraints on moral reasoning are surveyed.
Extreme value theory
On the decay of awfulness with oftenness
Wherein the limiting shapes of probability tails are classified and the Pickands–Balkema–de Haan theorem is invoked, so that excesses over high thresholds are modeled by the generalized Pareto law.
Algebraic probability
If you liked it then you prob’ly put a ring on it
Wherein an expectations‑first approach is presented, random variables being taken as primitives, measure spaces being dispensed with, and noncommutative (free) probability and convolution/transition semigroups being treated algebraically.
Marine biology and ecology
Wherein the midwater depths are treated, it is reported that ninety‑five percent of the world’s fish are hidden in the mesopelagic, and their vast biomass and role in climate dynamics are noted.
Gaussian processes
Wherein Gaussian processes are presented as probability laws over functions on domains such as R^d, being specified by their mean and covariance kernel, and being employed in regression and spatial inference.
Stochastic differential equations
Wherein stochastic dynamics are prescribed by integral equations driven by Brownian or Lévy noise, solutions are characterized pathwise and via generators and martingale structures, and differentiation of the noise is not defined without extra calculus.
Backward stochastic differential equations
Wherein a connection to a nonlinear Feynman-Kac representation is elucidated, viscosity solutions of the associated PDEs are invoked, and applications to stochastic optimal control and cost functionals are outlined.
Learning on tabular data
Wherein tabular data is presented as the common substrate for models, and gradient boosting machines are recommended as the practical recourse, while neural PFNs are noted for recent in‑context feats.
Random-forest-like methods
An optimally-weighted average of randomly stopped clocks is never far from wrong.
Wherein ensemble procedures composed of many weak decision-tree learners are treated, their aptitude for tabular data with minimal preprocessing and apparent self-regularizing behavior is recorded, and implementations such as XGBoost and LightGBM are surveyed
Optimal transport metrics
Wasserstein distances, Monge-Kantorovich metrics, Earthmover distances
Wherein the Wasserstein distance is introduced as a metric on probability measures, p=1 duality is noted, and an entropy‑regularised Sinkhorn variant is presented for fast computation on histograms.
Energy based models
Inference with kinda-tractable un-normalized densities
Wherein inference for undirected graphical models is framed as optimization of an energy function, and local gradient descent toward more probable configurations is depicted.
Ecology
Wherein a committee role for statisticians working in ecology in the local statistical society is held, a plate of dining birds is presented, and bibliographic references are enumerated.
What is your Sydney housing endgame?
Wherein a modest plan is proposed for a Sydney communal housing project, wherein mixed private and shared rooms are to be pursued, with emphasis on low per‑capita cost, nearby parks and simple governance.
Isotropic random vectors
Wherein isotropic random vectors are considered, multivariate Gaussian samples are normalized to produce uniformly distributed unit vectors, and axial marginals are deduced to have variance 1/d.
Neural net kernels
Wherein the covariances of infinite‑width neural nets are shown to depend only on input norms and their interangle θ, producing closed‑form Erf and arc‑cosine kernels expressed via arcsin and J_n(θ).
Randomized low dimensional projections
Wherein random low-dimensional projections of high-dimensional datasets are examined, and it is observed that, under mild conditions, empirical projected measures are shown to converge in law to Gaussian distributions
Burning bootable USB drives, SD cards etc
Wherein the competing tools for burning bootable media are presented, POSIX dd is invoked as sufficient, and examples such as Rufus on Windows and Etcher are enumerated, with writable persistent drives being noted.
Transforms of random variates
Wherein a nonlinear map is examined by a second-order Taylor expansion, and, for Gaussian inputs, the transformed mean and covariance are given in terms of the map’s Jacobian and Hessian, with unscented approximations noted
Cross validation
Wherein simulated out-of-sample folds are employed to select regularization strength for sparse regression, computational costs are outlined, and risks of information leakage in cross-methods are observed.
Complexity of markets
Computation in economic mechanisms
Wherein the computational limits of markets are examined, and the question of how market mechanisms are tasked with NP‑hard optimization problems like central planning is considered, with reductions to decision problems presented.
Deep Gaussian process regression
Wherein a layered construction of Gaussian process models is presented as a statistical layer cake, and an approximation via dropout from which neural‑network ensemble interpretations are obtained is explored.
Infinite width limits of neural networks
Wherein the infinite-width asymptotics of single-hidden-layer networks are shown to yield Gaussian-process limits under iid Gaussian weights, and kernel/NTK viewpoints and implications for implicit regularisation are surveyed.
Computational symbolic mathematics
Wherein non-commutative algebraic manipulations are examined via tools such as Cadabra, SymPy and Singular, and the nascent application of neural‑net methods to symbolic tasks is remarked upon.
Compressing neural nets
pruning, compacting and otherwise fitting a good estimate into fewer parameters
Wherein methods for reducing neural net size are catalogued, including pruning and lottery-ticket discovery, LassoNet-style feature selection with a skip layer, and quantization for deployment on edge devices.
Spectral factorization
Wherein a subcategory of Wiener–Hopf methods is presented, and spectral factorization of convolution kernels on the real line is treated as a procedural reduction to analytic factors.
Wiener-Hopf method
Righteous hack for certain integral equations
Wherein the technique for solving convolution-type integral equations is presented by factorizing kernels in the complex plane, and problems on half-axes are reduced to boundary-value formulations via analytic continuation and Fourier transforms.
Fourier transforms
Wherein the Fourier transform is considered for radial functions and windowed signals, Gaussian examples are exhibited and connections to the fast Fourier transform are indicated.
LaTeX Installation
Wherein the reader is guided through minimal TeX installs and alternatives such as Tectonic, and the practice of adding needed packages via tlmgr or TinyTeX’s auto‑install is delineated to save disk space.
Path integral formulations of SDEs
Feynman path integrals, esp for stochastic processes
Wherein path integral methods are applied to stochastic differential equations, the Onsager–Machlup action is formulated for Fokker–Planck dynamics, and distinctions between Itô and Stratonovich discretizations are examined.
Cats
Wherein the origin of multicolour pelage is disclosed, the orange locus is located on the X chromosome, and thus tortoiseshell or calico patterns are shown to indicate a female cat, via X‑inactivation during development.
Differentiable model selection
Wherein hyperparameters are tuned by backpropagating validation gradients through entire training runs, and learning-rate and momentum schedules are adjusted via hypergradients while initial weights are left untreated.
Prediction processes
Some kind of weird time series formalism
Wherein Cosma Shalizi’s proposal to unify ideas by means of chains with complete connections is set forth as a formal lens on prediction processes, and links to predictive processing of mind are noted.
Dynamical systems via Koopman operators
Composition operators, Dynamic Extended Mode decompositions…
Wherein the dynamics of observables are cast as evolution under an infinite-dimensional linear Koopman operator, its action mapping measurements forward in time and enabling spectral study of trajectories.
Kernel zoo
What follows are some useful kernels to have in my toolkit, mostly over \(\mathbb{R}^n\) or at least some space with a metric. There are many more than I could fit here, of…
Colour
Wherein the nature of colour is surveyed as a problem of dimensionality, and the reliance on four retinal photoreceptors in compressing an infinite spectrum is described alongside reproduction difficulties for print and RGB.
Generically approximating probability distributions
Wherein several methods of approximating probability laws are exhibited, Edgeworth expansions and kernel, empirical, and variational approximations are presented, and closeness is measured in probability metrics.
Assorted laws and paradoxes
Mostly-eponymous laws I would confound without this note
Wherein assorted laws and paradoxes are assembled and exemplified, a compendium is presented noting that Goodhart is sometimes confused with Godwin and that Moore’s law tracks transistor doubling.
Determinantal point processes
Wherein determinantal point processes are herein presented as point processes whose joint intensities are given by determinants of a kernel, and their repulsive sampling is noted for low-discrepancy quadrature.
Technopoetics
Wherein the interplay of code and verse is examined, and a range of practices is documented, from browser-based interactive fiction and Poemage visualizations to a poem encoded into bacterial DNA.
Statistics, computational complexity thereof
Wherein statistical inference is examined under fixed computation and memory budgets, the impact on Gaussian process methods is illustrated by O(N^3) scaling constraints, and limits on model expressivity given step and storage constraints are considered.
Matrix measure concentration inequalities and bounds
Wherein spectral-norm deviations for sums of independent Hermitian random matrices are bounded via matrix Chernoff, Bernstein and Efron–Stein inequalities, and Gaussian cases with Schatten-p estimates are given.
Measure concentration inequalities
The fancy name for probability inequalities
Wherein the notion of measure concentration is surveyed, classical inequalities are presented, and extensions to matrix concentration and high‑dimensional sums are exhibited with a corral metaphor.
Sauna
Wherein Australian sauna culture is surveyed and the public North Sydney Olympic Pool sauna is reported closed until 2023, and private venues and purchasable infrared cabins are catalogued.
Mind reading by computer
Wherein the bluntness of fMRI instruments and the reliance on subject-specific calibration are set out, and the extent of progress toward real-time decoding without priming or pretraining is reported.
Learnable memory
Wherein the question of how learning algorithms store and retrieve memories at inference is posed, how such memories are distinguished from weight storage is considered, and learnable longer-term transformer memory via extended context windows is introduced.
Convolutional Gaussian processes
Wherein Gaussian processes are constructed by convolving white noise with spatially varying smoothing kernels, so that locality is enforced and non‑stationary covariances are induced by tuning the driving noise or kernel.
Gauss markov random fields
Precision vs covariance, fight!
Wherein Gaussian Markov random fields are represented as linear SDEs via Green’s functions, and stationary Matérn covariances are realised by all‑pole state‑space models.
Stochastic processes on manifolds
Wherein stochastic processes on manifolds are examined, and the influence of curvature on local diffusion and Brownian motion in coordinate charts is described, with attention to metric-dependent drift terms.
Learning covariance functions
Learning a family of covariances at once
Wherein the extension of covariance estimation to continuous index sets is described and parametric kernels are selected by maximising marginal likelihood for Gaussian processes, with compositions learned by gradient methods.
Frames and Riesz bases
Generalisations of orthogonal bases
Wherein the redundancy of frames is exhibited and signal decompositions are allowed to be nonorthogonal, enabling denoising and compressive sensing methods via restricted isometry properties.
Causal inference in the continuous limit
Wherein graphons are examined and are found insufficient to capture covariate-driven spatial or temporal proximity for fields of continuously indexed variables, and light-cone intuitions are considered.
Stability in linear dynamical systems
Wherein systems are considered stable when they avoid super‑polynomial explosion, poles on the unit circle are allowed, and reparameterisation techniques are employed to enforce stationarity in multivariate models.
Chaos expansions
Polynomial chaos, generalized polynomial chaos, arbitrary chaos etc
Wherein a method for expressing random quantities as orthogonal polynomial series with respect to a chosen germ distribution is presented, and truncation is indicated as a practical means to track propagation of uncertainty.
Office software
Wherein various office suites are surveyed, alternatives to Microsoft are presented, including LibreOffice, OpenOffice, FreeOffice and cloud options, and Linux compatibility is noted.
Make your own podcast why not
Wherein common podcasting services such as Anchor and Audioburst are catalogued, transcription-assisted editing via Descript is described, and legacy loudness normalization by the Levelator is noted.
Online collaboration
Wherein the shift to virtual co-working is described, and the loss of incidental workplace interactions, effects on early‑career recognition, communicative silences, and practices like gather.town co‑working are examined.
Integral transforms
Wherein integral transforms are presented as tools for solving integral equations and partial differential equations, and the Laplace transform is expressed in probabilistic form as the expectation E[e^{-sX}].
Feynman-Kac formulae
Wherein the linkage of Feynman–Kac formulae to central limit theorems for sequential Monte Carlo filters is presented, connections to backward SDEs and the Fokker–Planck equation are noted, and canonical expositions by Del Moral and Doucet are cited.
Miscellaneous nonstationary kernels
Wherein nonstationary kernels, constructed by methods other than warping, are presented, with explicit basis expansions and spatially varying-parameter formulations being exemplified and their historical roots traced to Jun 2008 and Fuglstad 2013–2015.
Warping of stationary stochastic processes
Wherein bijective input warpings are presented to render kernels stationary, and local positive‑definite lengthscale matrices are employed to produce nonstationary covariances used in Gaussian process regression.
Running a secure server
and other self-hosting madness
Wherein a hardened operating system is recommended, firewall rules are applied, and TLS provisioning via Let’s Encrypt or Cloudflare is described so development servers are protected against casual port‑scans.
Bayesians vs frequentists
Just because we both get the same answer doesn’t mean neither of us is wrong
Wherein Bayesian and frequentist doctrines are compared as competing inferential frameworks, Freedman’s inconsistency result is mentioned, and the computational hardness of exact Bayesian updating is noted.
Auditory features
descriptors, maps, representations for audio
Wherein differentiable, invertible, and psychoacoustically informed audio descriptors are surveyed, and the prevalence of noninvertible MFCCs versus raw-audio neural features that recover Mel-like bands is noted.
Positive (semi-)definite kernels
Wherein kernels are presented as covariance functions of stochastic processes, and it is noted that positive semidefiniteness is required so that all finite linear combinations have nonnegative variance.
Nowhere to hide
A catalogue of terrifying surveillance methods I encounter
Wherein the reader is informed that the reach of cheap surveillance is shown by instruments that see through walls with Wi‑Fi and identify individuals by heartbeat from hundreds of metres.
Why does deep learning work?
Are we in the pocket of Big VRAM?
Wherein the role of stochastic gradient descent is examined as a statistical‑mechanics‑like process, the interplay of overparameterization with SGD is shown to permit efficient finding of global optima, and approximation is observed to favor depth over width
Generative adversarial networks
Wherein two networks are trained adversarially, a generator and a critic, and the procedure is framed as an optimal-transport problem using a Wasserstein loss to permit likelihood-free simulation.
Garbled highlights from NeurIPS 2020
Wherein the conference’s virtual workshops and papers are collated, and an emphasis on causal discovery, continuous‑time Neural ODE models, and climate‑focused ML workshops is recorded.
Sun protection
Wherein a preference for the Eclipse hoodie-shawl as a cycling garment is recorded, and sun-protective apparel such as Sun Bella is noted for use with bikes and parasols during rides and outdoor passage.
Random embeddings and hashing
Wherein random embeddings and hashing are examined, and kernel approximation via random projections is proposed as a concrete application, with links to Johnson–Lindenstrauss and Cover’s theorem.
Randomised regression
Wherein random embeddings of predictors are employed to reduce high-dimensional regression via low-dimensional projections, and differences in noise handling from compressed sensing are noted, with attention given to dependent time-series data.
Distributed consistency
Getting stuff down in crowds of computers
Wherein the coordination of remote processes by message passing is described, and conflict‑free replicated data types are cited as a commonly employed, intricate approach to achieving eventual consistency.
Distribution regression
Wherein a regression is presented in which the covariate is a probability distribution observed only via finite samples, and the response is modeled as a function of that distribution plus random error.
Probabilistic spectral analysis
Wherein the nonstationary spectrum of audio speech is modeled by Gaussian process carrier waveforms multiplied by a spectrogram that is learned via nonnegative matrix factorisation with GP priors from recordings.
Tensor regression
Wherein the extension of linear regression to multilinear forms is presented, and the use of tensor decompositions for parameter reduction and the Tensorly library as a practical implementation are noted.
Julia arrays
Wherein Julia arrays are treated as 1‑indexed by convention but are shown to permit arbitrary indices, iteration and slicing methods are outlined, and column‑major storage with broadcasting for GPUs is noted.
Observability and sensitivity in learning dynamical systems
Parameter identifiability in dynamical models
Wherein the limits of parameter recovery are examined and transfer functions are identified as the actual learning targets, and sensitivity of observations is quantified via local and global gradient analysis
Markov Chain Monte Carlo methods
Wherein sampling from approximate distributions by reweighting is outlined, importance sampling is explicated, and its application in particle filters for sequential state estimation is presented, with a textual reference to Art Owen.
Julia, testing and packaging
Wherein package scaffolding via PkgTemplates and experiment provenance via DrWatson are set out, CI and test coverage scaffolds are provided, and code versions are attached automatically to simulations.
Efficient factoring of GP likelihoods
Wherein GP likelihoods are shown to be factorised by combining sparse inducing points with variational approximations, so that high-dimensional expectations reduce to functions of univariate Gaussians.
Localized Gaussian processes
Wherein the value at a location is predicted from only the nearest observations, and the screening effect is shown to depend on algebraic high‑frequency decay and limited variation of the spectral density.
Variational Gaussian Process regression
Wherein a variational approximation to Gaussian Processes is presented, employing a sparse inducing-point formulation and Kullback–Leibler divergence minimization for scalable inference.
Audiovisuals
Synesthetic and other cross-media audio stunts
Wherein the co-generation of imagery with sound is presented, and Amazon field experiments are adduced to exemplify crossmodal production and to illustrate temporal mapping between audio and image.
Time
Wherein the modern bewilderment of chronology is surveyed, the peculiar intrusion of leap seconds upon civil timekeeping is examined, and Unix epoch and calendar oddities are catalogued.
Stan
The flagship Bayesian workhorse
Wherein Stan is presented as a probabilistic programming language implementing Hamiltonian Monte Carlo for Bayesian inference, and limitations regarding discrete parameters, positive‑definite matrix domains, and absence of a foreign‑function interface are noted.
Julia IO
Wherein the IO ecosystem for Julia is surveyed, and bindings for common scientific formats such as CSV, HDF5, NetCDF and Arrow (Feather) are catalogued, while Protobuf and database‑backed BigArrays are noted.
Fintech, assorted
Wherein Australian fintech ventures and retail share‑trading platforms are surveyed, regulatory shifts and platform growth are noted, and connections to historical exchange practice are traced.
Convenient Razer
Wherein the Razer Blade’s Linux compatibility is examined and it is noted that firmware updates are obtainable only by dual-booting Windows, while various peripheral and suspend peculiarities are catalogued.
Alternative file managers
Wherein the Ubuntu default file manager is supplanted by alternatives, fman is presented for its command‑palette and EUR18 licence, and Krusader is described as a feature‑rich dual‑pane KDE option.
AutoML
Wherein a controller neural net is described as proposing child architectures for evaluation, and hyperparameter tuning is automated by Bayesian search and ensemble construction, all is presented in sober detail
Monte Carlo optimisation
Wherein Monte Carlo optimisation is presented as being performed via Markov Chain Monte Carlo with annealing to traverse multimodal objective landscapes, and convergence checks are specified.
Splitting simulation
Wherein the technique is described, attention is narrowed to the high-probability region of an intractable distribution by partitioning the state space and alternating local recombination and resampling.
Webcams
Wherein webcam software and Linux compatibility are surveyed, and the practical detail that one popular application is found to record silent videos is reported, while the substitution of a real camera is proposed.
Quantitative risk measurement
Mathematics of actuarial and financial disaster
Wherein the measurement of financial peril is treated, and quantiles such as Value‑at‑Risk and Expected Shortfall are presented as tools for assessing tail losses, with notes on climate applications and scheduling.
Graph computation
Wherein graph computations are framed via sparse adjacency matrices and Laplacians, and engines for spectral methods, clustering, large-scale iterative processing, and pattern-mining distinctions are surveyed.
Filter design, linear
Wherein discrete-time linear filters are considered, the z-transform and bilinear mapping are invoked, Bode plots are employed, and sampling relations are addressed to prescribe digital IIR and FIR responses for software implementation.
Gaussian process quantile regression
Wherein a procedure for predicting a chosen conditional quantile is presented, under an asymmetric pinball loss, and Gaussian process priors are used to model the latent regression function for uncertainty-aware estimates.
Independence, conditional, statistical
Wherein the graphoid axioms are laid out and modern nonparametric tests such as Chatterjee’s ξ and kernel conditional independence methods are described, and connections to graphical models and model selection are sketched.
Statistics of spatio-temporal processes
Wherein spatial processes evolving through time are considered, and a discretised graphical-model perspective with sensor observations and ensemble Kalman filtering for state and parameter inference is presented.
Data dimensionality reduction
Wherein methods for reducing predictors are surveyed, and the learning of similarity metrics and manifold embeddings is presented as a means to produce low-dimensional summaries that aid indexing and inference
Variational autoencoders
Wherein the VAE is presented as a probabilistic autoencoder in which a low-dimensional latent space is imposed and a reparameterization trick is employed to permit gradient-based learning.
Neural nets
Designing the fanciest usable differentiable loss surface
Wherein the revival of deep learning is described, and dependence on incremental algorithmic tweaks and SIMD GPU hardware for training many-layered models on vast datasets is noted, with recurrent and convolutional variants mentioned.
Homo ludens
Wherein the play of humans is examined as a social force on the dance floor, and the politicized rituals of rave, gentrification and commodified dissent are traced as manifestations of leisure’s power.
Live looping music software
Wherein live looping software is surveyed and catalogued; ageing open-source and commercial tools are described in deadpan detail, and a concrete technical note is recorded about support for loops up to one hour in length.
Causal graphical model reading group 2020
An introduction to conditional independence DAGs and their use for causal inference.
Wherein the preliminaries of causal DAGs are recounted, do‑calculus and the back‑door criterion are introduced, and a case study on causal Gaussian‑process regression is examined.
Emulators and surrogate models via ML
Shortcuts in scientific simulation using ML
Wherein emulators are reduced physics simulations by machine learning, and dropout ensembling is employed to yield quasi‑Bayesian predictive uncertainty for surrogate optimisation in experimental design.
Generative adversarial networks for PDE learning
Wherein generative adversarial networks are employed to infer solutions and operators of partial differential equations, and fluid-flow dynamics are illustrated through density-to-velocity reconstructions.
Ethnomusicology in the technologised, globalised world
Global bass, World Music 2.0, folkwaves etc
Wherein the commodification of diasporic pop forms is traced through digital networks and copyright disputes, and Jamaican street-dance practices are held up as a case of technological and legal encounter
Mathematics of textiles
Wherein braid theory, jacquard looms, and lace are examined, connections to topology and computer science are traced, and the coding of fishnet stocking patterns for industrial jacquard weaving machines is described, with references to digital embroidery tools
Minimum description length
And other takes on learning-as-compression
Wherein a formalisation of Occam’s razor is presented, a link to Bayesian marginal likelihood for model selection is noted, and the practical uncomputability of MDL in neural‑net settings is observed.
Remote Desktop
Business model: Uber for pixels
Wherein remote graphical terminals are catalogued by protocol—SSH/X11, NX, VNC, RDP—and the multiprotocol Linux client Remmina is noted as a practical nexus; browser and SPICE variants are also mentioned.
Plotting stuff in julia
Wherein it is observed that Julia plotting defaults to SVG, producing large files and rendering cost, so raster PNGs are preferred for exploratory work, and PGFPlotsX is noted for native LaTeX support.
Combinatorics of note
Wherein combinatorial connections between algorithmic complexity and quasi-Monte Carlo are considered, and Jörg Arndt’s Matters Computational is cited as a referenced resource for further computational examples.
Networking stunts
Wherein techniques for diagnosing local network behaviour are described, including how the process bound to a port is identified via lsof and ss, and tools for measuring and throttling bandwidth are noted.
Citation management
Wherein the modern travails of scholarly citation are surveyed, and Zotero is adopted as the principal tool for importing, syncing, and storing PDFs, while BibLaTeX and pandoc are employed to render bibliographies.
Low code development
Wherein low-code development is presented as a practice in which spreadsheets and graphical flow-based programming are cited as common interfaces, and atypical visual models are examined.
Task launchers
Wherein various app launchers are surveyed and it is observed that Electron-built candidates are heavy on RAM, Python-extension ecosystems are cataloged, GNOME’s built-in launcher is affirmed, and choice fatigue is reported.
Financial markets
Descriptive engagement with reality of
Wherein modern financial markets are examined, and the rise of high-frequency trading is noted as a phenomenon that can consume liquidity while venture-capital exit pressures are observed to reshape firm financing and intermediation costs
Economics of insurgence
Wherein the modern insurgency is surveyed as a businesslike enterprise, with MBAs and business analytics applied to self‑organising violence and social‑media mobilization, and freelance criminal networks are catalogued.
Debugging, profiling and accelerating Julia code
Wherein Julia’s debugging, profiling and speed tactics are surveyed, REPL-only debuggers and Profile/ProfileView are noted, and explicit LRU-style in-memory caching patterns are recommended for heavy computations.
Firms
Wherein the reasons firms arise, their Zipf-like size distribution, and the absence of a single economic monopolist are considered, and firms are treated as proto-AIs shaped by internal bookkeeping
Deep fakery
Wherein the coming ubiquity of video falsifications in weaponised social media is surveyed, and the influence of audio forgeries on the cost curve of reality fabrication is noted.
Encrypting, signing, verifying stuff
Wherein practical cryptographic tools and usability issues are surveyed, and Keybase’s social-identity verification and GUI conveniences are presented as an alternative to manual GPG workflows for encrypting, signing, and verifying.
Productivity schemes
Wherein the necessity of slack time is examined, quantified-self methods are surveyed, optimization is depicted as calcification, and the avoidance of email and social networks is urged.
Installing Julia
Wherein the Julia binary is placed on macOS by creating a symbolic link from /Applications/Julia-1.1.app/Contents/Resources/julia/bin/julia to ~/bin/julia, and alternative installs are noted.
Inference on social graphs
Heterogeneous media and controls
Wherein egocentric sampling is examined, the friendship paradox and majority illusion are invoked, and a graph‑Laplacian method is proposed to infer transmission rates across networks from partial observations.
Long memory time series
Wherein the role of Hurst exponents and fractal models is examined, and links to 1/f noise, fractional Brownian motion and branching processes are indicated as generic mechanisms for long memory
Databases for realtime stuff
Wherein real‑time databases are surveyed and the suitability of in‑memory stores such as Redis for heavy write traffic and millisecond updates is noted, with ring‑buffer RRD and streaming Materialize also described.
Link rot, mitigating
Wherein link decay is described and automated detection and preservation are recommended, including use of the command‑line checker hyperlink with domain skips and ArchiveBox to save HTML, PDF and WARC copies
How to reduce government spying on you
Wherein methods to limit state surveillance are outlined, and the use of anonymous document submission via SecureDrop and hardware precautions such as USB condoms and hardened OSes is described.
Natural gradient descent
Climbing slower on the tricky bits
Wherein the ordinary gradient is found inadequate and the direction of steepest descent is defined by the Fisher information, so gradients are rescaled to respect KL divergence between probability distributions.
Malliavin calculus
Wherein a calculus is presented by which derivatives of probability densities for stochastic processes are defined on Wiener space, and consequences for stochastic differential equations are inferred.
Variational inference
On fitting the best model one can be bothered to
Wherein variational inference is presented as an optimisation of an approximate density, the ELBO using KL divergence is introduced, and reparameterisation tricks are described.
Lévy stochastic differential equations
Wherein Lévy-driven stochastic differential equations are examined, sparse stochastic process sampling theory is invoked and a Poisson-driven example is exhibited, and Malliavin calculus is mentioned as a potential tool.
Screen capture and screen casting
Wherein methods of capturing and casting screens are surveyed for Linux use, with particular attention paid to OBS, Kazam and Narakeet, and to Wayland-related limitations affecting live broadcast workflows.
Javascript apps
Wherein the making and testing of JavaScript apps is treated, and the possibility of offline operation via ServiceWorker, local development on a localhost server, and alternative runtimes like Deno and Tauri is recorded.
MAPLE
Wherein the symbolic system Maple is examined, is noted to employ an imperative emphasis rather than a purely functional style, and is demonstrated in examples to handle transforms of random variables.
Graphic novels/comics etc
Stuff I won’t admit to reading when there are deadlines
Wherein an inventory of graphic novels and comics is presented, with numerous online serials and accompanying RSS feeds enumerated for archival and ongoing access, and stereoscopic illustrations cited.
Directed graphical models
Wherein directed graphical models are described as DAGs, causal interpretations are considered, Simpson’s paradox is exhibited, and a Julia BayesNets package for reasoning is noted.
New-wave jigsaw puzzles
Wherein an index of novelty jigsaw designs is presented, including colour-changing and illusionary puzzles from makers such as New York Puzzle Company, and a practice of sending extraordinarily difficult sets to one’s mother is recorded.
Adaptive Markov Chain Monte Carlo samplers
Wherein adaptive samplers are presented, and transition parameters are made dependent on the chain’s history, so proposal kernels are updated online and pilot runs are proposed to assess convergence
Tuning an MCMC sampler
Wherein the sampler is tuned by pilot runs and by online adaptive schemes, suspect pilot samples are discarded, and expected squared jump distance is proposed as the optimisation criterion for optimal mixing and reduced rejection rate.
Empirical estimation of information
Informing yourself from your data how informative your data was
Wherein empirical estimation of information is undertaken, and a Monte Carlo device for approximating KL divergence by sampling from q while evaluating p and q is explicated, rare low-frequency observations being shown influential.
C++
Wherein the rewriting of Python inner loops in C++ for modest CPU savings is undertaken, and notable libraries such as Boost and OpenCV are enumerated, while direct use of Tensorflow is mostly eschewed.
The cross entropy method
Wherein the cross entropy method is applied to optimise a proposal distribution for rejection sampling within a Monte Carlo simulation context, and a worked example is constructed for eventual inclusion in Wikipedia.
Mixture models for density estimation
A method of semi-parametric density estimation.
Gym togs
On getting fancy activewear of low impact to my pocket and the planet
Wherein activewear is examined with special emphasis on recycled, low‑impact gear and Wilderness Wear’s Tasmanian provenance is noted, shoes are filed elsewhere, and sun protection is considered.
Waste, and the late capitalist industrial metabolism
Wherein the peculiar commerce of bottled water and a grassroots compost network are described, and it is noted that single-use cups are charged at roughly the same cost as bottles, notably on some campuses.
Matching and weighting
Making the optimal beverage from the fruit life gave you
Wherein multilevel regression and poststratification is presented as a technique for correcting survey non‑response bias, and its formulation with structured priors and Bayesian implementations is outlined.
DIY computer radio
Wherein the reader is informed that a modest computer is employed as an FM transmitter via the Raspberry Pi GPIO clock output, and low-bandwidth audio networking is surveyed by enthusiasts.
Garden hacks
Wherein the avoidance of urban fruit cultivation is explained by elevated lead in the soil, and the adoption of no‑dig portable beds and community compost‑sharing is described.
Supercollider
Wherein the paired languages for live interactive music are described, the DSP backend scsynth is noted as operating separately and being driven by the ageing frontend via the 1999 OpenSoundControl network protocol.
How to reduce criminals spying on you
Wherein the reader is informed that attackers exploit reused passwords and can intercept two‑factor tokens via man‑in‑the‑middle tools like Evilginx, and basic device hardening is advised.
Analysis/resynthesis of audio
Wherein audio is analyzed by machine listening, sparse low-dimensional features are extracted, stochastic models are fitted and simulated for resynthesis, and concatenative mosaicing from a learned sparse dictionary is employed while psychoacoustic cost functions are considered
Linux hacks for the command-line
Intermittently needed cheat codes
Wherein common Linux command-line enigmas are catalogued in a prose of practical admonition, and the use of xdg-open to open GUI files from the shell is demonstrated alongside kernel module rebuild notes.
Queueing
The mathematical field whose major result is enraging you about call centres
Wherein Kingman’s approximation for waiting time is presented, and the expected queue delay is expressed in terms of utilization, coefficients of variation of arrivals and services, and mean service time.
Hygiene, empirical
Wherein common practices of handwashing, masks and surface measures are examined, and handwashing routines and particulate mask efficacy are empirically considered as means to reduce infectious disease transmission.
Epidemics
Wherein attention is directed to the mechanics of disease spread, with particular examination of COVID‑19 and its ties to global trade networks, and empirical contagion models are considered.
Teaching computers to write music
Wherein the training of machines in musical composition is described, and MIDI sequence generation with TensorFlow’s Magenta, piano-roll representations, polyphonic and RNN-based models alongside maximum-entropy approaches are examined.
Graphic design for the vexed
Wherein colour theory and practical tools such as Figma, Canva, and Hatchful are surveyed, and diagramming, UX, image editing, and website cheats are furnished as applied aids.
The surveillance society
Wherein real‑time facial and object indexing by corporations is described, and insurers’ harvesting of ambient data is shown to be used to score, discipline, and alter employment and coverage prospects.
Encrypted filesystem on linux
Wherein whole-disk and per-user encryption methods are surveyed, Ubuntu’s adoption of LUKS as the default is noted, and the practicalities of double-login prompts and swap-held keys are described.
Order statistics
Wherein the ranks of a sample are defined, and a representation for sums of the top-k of iid exponential variates via quantile transforms is exhibited, and connections to maxima and copulas are sketched.
DIY internet infrastructure
Wherein various low-cost communication schemes are surveyed, and a Raspberry Pi is described as being used to convert streaming audio into FM broadcasts for local distribution.
Tool discovery
Settling upon an adequate gizmo to identify other adequate gizmos
Wherein various sites for software selection are surveyed, crowdsourced alternative listings such as Alternativeto are presented alongside category-organized services like Product Hunt, and Yourstack is noted as web-dev oriented
Faking being on social media
and other parts of the internet too boring to countenance
Wherein the illusion of social-media presence is maintained by automated means, Facebook Messenger activity is simulated via APIs, bots and bridges such as Matterbridge, and commercial bot services are engaged.
Restricted isometry properties
Restricted isometry properties, a.k.a. uniform uncertainty principles (E. Candès and Tao 2005; E. J. Candès, Romberg, and Tao 2006), mutual incoherence (David L. Donoho 2006; …
Diff/merge tools
Wherein various tools for comparing and harmonising folders and files are surveyed, and both GUI merge utilities and recursive unpacking diff tools that transform binaries into readable forms are catalogued.
Kernel approximation
Wherein kernel feature maps are approximated by explicit random embeddings, attention being given to translation‑invariant kernels and computational trade‑offs using random Fourier features and Nyström methods
System monitoring
Wherein a Raspberry Pi VPN router is observed to be overheating and failing to load iptables rules, and simple time‑series recording with RRD or tools like netdata is considered for diagnosis.
Schwiizertüütsch
Wherein the language of Switzerland is described as part of the Alemannic dialect zone, and peculiar local lexemes such as the word for kitchen cupboard are noted, while etymologies are catalogued.
Switzerland
Wherein a former residence amid Alpine bath mountains is described, and the author’s dabbling in Swiss German is recorded, while links to local cultural and environmental organizations are noted.
Effective sample size
Wherein the reduction of independent information by serial autocorrelation is formalized and an adjusted sample count is defined, with Markov chain autocorrelations yielding a scaled effective sample size.
Bias reduction
Estimating the bias of an estimator so as to subtract it off again
Wherein a reduced-bias M-estimation method is described, in which analytical adjustments that depend only on first and second derivatives are applied and bias is lowered without simulation or expectation computation.
HTML for haters
Wherein the barest modern HTML5 and CSS are prescribed for minimalist sites, accessibility tips and Flexbox responsive patterns are espoused, and pragmatic browser support is limited to common users.
Learnable indexes and hashes
Wherein learnable indexes and hashes are surveyed, and learned bloom filters alongside k‑nearest‑neighbor methods via Neighbourhood Components Analysis are examined for their roles in similarity search
Microsoft Azure cloudydoodle numberpants crunchery
Wherein Azure’s data-science offerings are surveyed, and FPGA acceleration, GPU VMs, VS Code integration, serverless Functions, and an ML Workbench quickstart are cited for varied workflows.
Selling uncertainty
On the marginal price of ignorance
Wherein the sowing of doubt is chronicled, tracing how industries and hired experts replicate the tobacco playbook to obscure harms such as pesticides and climate risk.
Concurrent programming
Threads, locks, cores, conditions
Wherein threading and locking are examined through a sequence of example programs, condition variables are explained, and POSIX threads are introduced as the terminology and patterns to be learned.
Sunda
Wherein the Sundanese region is delineated, its distinct script and the ceremonial kujang blade are noted, the Pranatamangsa astrological calendar is mentioned, and the Sunda Wiwitan faith is described beneath Hindu and Islamic layers.
Cepstral transforms and harmonic identification
Wherein the spectrum is represented via a log link to the power spectrogram, cepstral methods are used to expose long correlation lags, and a connection to MFCCs in machine listening is noted.
Yak shaving
Wherein a tendency to pursue nested errands is examined, its MIT coinage origin is noted, and the question of whether simpler actions can avert cascading failures in complex systems is posed.
R Shiny
Statistics through the internet
Wherein an interactive R webapp generator is described, containerized deployment via Docker is outlined, and browser frontends to R backends with DT/DataTables tabular display are noted.
Random change of time
Wherein a.e. continuous monotonic random changes of index are examined, and Ogata’s time‑rescaling via compensators is shown to convert point processes to unit‑rate Poisson processes, while links to the Lamperti representation for continuous‑state branching are noted.
Cheap single board computers
Wherein cheap single board computers are examined as disposable miniature machines used to run separate apps for improved security, and practical notes on Raspberry Pi 4 thermals and firmware are provided.
Branching processes
Wherein continuous-state variants and Hawkes-type self-exciting point processes are presented, and the problem of parameter estimation from finitely sampled continuous branching trajectories is considered.
Infinitesimal generators
Generators of the transition semi-group, connection to Kolmogorov forward equations
Wherein infinitesimal generators are presented as the time‑derivative of the transition semigroup, are used to approximate short‑time evolution and are shown to yield differential operators that may involve infinitely many derivatives.
Poisson point processes
Wherein a process is presented whose inter-arrival times are exponential and whose counts on [0,t] are Poisson(λt), the conditioned bridge at t is shown Binomial(S,t), and hitting times are Gamma(ℓ,λ).
Psychoacoustics
A quick incomplete reference to pascals, Bels, erbs, Barks, sones, Hertz, semitones, Mels and whatever else I happen to need.
Differential privacy
Wherein a randomized coin-flip mechanism for eliciting private counts is described, and the recent selection of differential privacy for U.S. census data and practical libraries and tutorials are surveyed.
*-omics
Wherein the mapping of proteomes, genomes and phenomes is presented as networks, and statistical inference of control pathways using model selection and false‑discovery considerations is examined.
Learning in adaptive systems
On staring into scopophilic abysses
Wherein heuristics about learning in adaptive systems are set out, and the contrast between non-arbitrage constraints and conventional hypothesis tests is delineated, with scale effects in economics noted.
Switching to netlify
Wherein the blog is migrated from Pelican and GitHub to blogdown and Netlify, and interruptions, confusion, and aesthetic disorder are experienced, while the site is rendered sufficiently operational for present use.
Media metadata management and editing
Wherein media metadata management is surveyed and editing practices are described, with attention to transcoding workflows and linked notebooks for music analysis, PDFs, and image handling.
Wikis
Wherein a catalogue of collaboratively edited knowledge repositories and tooling is compiled, and markdown-oriented Outline, single-file TiddlyWiki, and MediaWiki-style server software are noted as practical options.
Trains
Wherein the author’s preference for rail travel is explained and the Victorian coinage gunzel, coined in the mid‑1970s to denote an obsessive tram or train enthusiast, is recounted.
Esoteric language zoo
Wherein the reader is guided through arcane tongues, from Iversonian APL kin and Whitney’s k and b to Brainfuck’s homomorphic encrypted implementation and INTERCAL’s COME FROM jest.
Information geometry
Wherein the study of statistical manifolds and learning on curved spaces is presented, and pointers to the Geometric Science of Information conference and to an Azimuth series on information geometry are provided.
FFMPEG
Wherein the multipurpose media tool is presented, and concrete methods for extracting audio and re-encoding outdated camera footage to modern codecs (libx264/libx265) via Homebrew builds are detailed.
Hawkes processes
Wherein the self‑exciting point process is described, its intensity is given as a base rate plus a past‑triggered kernel weighted by a branching ratio η, and time‑varying background is effected by top‑hat kernels.
Linux filesystem hacks
Wherein the care of filesystems is expounded, TRIM is recommended for SSD longevity and instructions are provided for adding exfat support to Ubuntu via apt install exfat-fuse exfat-utils.
Controllerism
Making thing happen by waving your arms about on stage
Wherein the methods of bending a controller to musical intent are described, and mappings from high‑dimensional instrument parameters to low‑dimensional controllers are examined via regression and copulas.
Permanental point processes
Wherein a point process is described whose intensity is given by the square of a Gaussian process, and whose likelihood is seen to involve a matrix permanent in its formulation.
Spatial point process and their statistics
Wherein the theory of spatial point processes is surveyed, and their use in earthquake modelling and latent Gaussian–driven Cox processes with pseudolikelihood-based inference is delineated.
Convergence of random variables
Wherein the distinctions between modes of convergence are examined, and the absence of any metric for almost-sure convergence on probability spaces is highlighted, while metrizable notions such as convergence in distribution and in Lp are contrasted.
Cherchez la martingale
Stuff about probability and orthogonality
Wherein the ubiquity of martingales in stochastic processes is expounded, and the classical double‑down gambling strategy is identified as a local martingale, while possible applications to limit theorems and estimators are suggested
Clojure
Wherein Clojure is treated as a library managed via Leiningen, its employment in generative art workflows is recorded, tooling and tutorials are catalogued, and small scripting projects such as fleck for shell automation are cited.
Local and networked UIs in Julia
Wherein browser-based Julia interfaces are described, and deployment pathways are noted, including packaging via Blink.jl so that Interact.jl apps are deployable to the IDE, a web server, or an Electron desktop.
Audio source separation
Wherein the decomposition of commercial recordings into stems is described, it is noted that scarce isolated-track corpora constrain training and that models are used to yield separate vocals, drums, bass, and accompaniment
Trusting information
Spinning swarm sensing from comment threads
Wherein methods for assessing others' trustworthiness are surveyed, and systems for proving identity and truth—via notary-signed artifacts, webs of trust, reputation graphs and blockchain proofs—are examined.
Bitwig
Wherein Bitwig is presented as a rebooted Live‑like DAW, and its open JavaScript and Java controller APIs are outlined, with platform‑specific controller script paths and native Linux support.
Sharing / gig economy
Wherein platform-mediated labour is chronicled as a new iteration of old economies, in which workers are classified as contractors, economic risk is transferred to them, and labour is micromanaged by smartphone apps.
MIDI
The near-adequate compromise for digital music that we are stuck with
Wherein MIDI is presented as a longstanding 7-bit messaging protocol, networked flavors such as RTP‑MIDI and ALSA are outlined, and General MIDI percussion key numbers are provided.
Tunings
Wherein microtonal practice is outlined and the eccentric Scala software is described as requiring arduous Linux installation and MIDI configuration to access its extensive library of scales and converters.
Machine listening
Wherein machine listening is treated as the application of statistical and feature-based machine learning to musical audio, and implementations such as LibROSA and Essentia are outlined.
Coupling in stochastic process
Wherein the coupling method is presented as a tool to bound Markov chain convergence and to construct exact and debiased Monte Carlo samplers, including Coupling From The Past and debiasing schemes.
ISMIR 2019
Wherein the twentieth congress is held in Delft, workshops on GAN-based music generation and waveform deep learning are presented, and novel source-separation systems alongside a dance-video corpus are showcased.
Phase retrieval
I’ve got the power. / Like the crack of the whip/ I snap attack/ Front to back
Wherein the reconstruction of lost phase is undertaken and iterative methods such as the Griffin–Lim algorithm are presented, and gradient-based Wirtinger flow strategies are described for practical recovery.
Webhooks
Wherein a generic internet interoperation standard is described, and a self‑hosted DIY tool called webhook by Adnan Hajdarevic is noted for being used to automate GitHub deployments and run listeners as system services.
Smartypants cities
Wherein ubiquitous sensors and automated controls are proposed to monitor and manage the built environment, municipal decisions are routed through continuous data streams, and urban movement is rendered legible.
Optimal control
Wherein the theory of linear quadratic regulators and Riccati equations is outlined, and recent approaches framing control as online regret minimization via gradient perturbation, with links to H∞ robustness and POMDPs, are recorded.
Generative art, creative coding, procedural design
Teaching my computer to make prettier mistakes than me
Wherein algorithmic artists are described as designed entities, techniques such as flocking and L‑systems are enumerated, and practical tools like Processing and SuperCollider are catalogued for hands‑on praxis.
Undirected graphical models
Wherein spatial Poisson and Bernoulli random fields for discrete multivariate sequences are considered, and Gibbs–Boltzmann inference methods for Markov random fields are outlined.
Gradient descent, Higher order
Wherein higher-order extensions of gradient descent are examined, and 3rd-order Halley–Chebyshev methods are noted to require tensors beyond Hessian matrices, rendering them costly in multivariate settings.
Ableton Live
The de facto standard for techno
Wherein Ableton Live is treated as a ubiquitous DAW, and its scripting avenues, including Max for Live and LiveOSC, are examined, while its project files are noted to be stored as gzipped XML.
Rhythm
Especially for generative music
Wherein the detection and construction of periodic patterns is examined through Kuramoto oscillators, autocorrelation structures and the manipulations of breakbeat cuts, and methods are outlined.
Delays and reverbs for audio processing
Wherein audio recurrence for music is considered, and stable MIMO delays are presented using orthogonal and unitary matrices to parameterize multichannel allpass feedback for echo-like textures.
Frequentist consistency of Bayesian methods
TFW two flawed methods for understanding the world can agree with at least each other
Wherein the alignment of Bayesian procedures with frequentist consistency is examined, posterior concentration and predictive convergence are considered, and failure of the Bernstein–von Mises in infinite dimensions is noted.
Discrete time Fourier and related transforms
Also, chirplets, z-transforms, chromatic derivatives…
Wherein fast inversion of chirp z-transforms is shown to be as tractable as FFTs, and complexity, timings, windowing, and chromatic derivatives for operators on discrete-time series are examined.
Density estimation
Especially non- or semiparametrically
Wherein a distribution is sought directly, nonparametric densities are treated as function-approximation problems, and practical issues such as kernel bandwidth selection and divergence choice are examined.
The tidyverse
Wherein is described a compact ecosystem in R, centered on dplyr and functional conventions, and is accompanied by tools for reshaping data such as pivot_long, and is exemplified by Hadley Wickham’s writings.
Bio computing
Wherein living organisms are employed as logic gates and general computing devices, and initiatives such as Microsoft’s Station B are noted as being pursued toward applied biological computation
Zeros of random trigonometric polynomials
Wherein is considered the expected number of real zeros of trigonometric polynomials with nonidentically distributed coefficients, and connections to determinantal point process models are indicated.
Generalized Galton-Watson processes
Wherein a discrete-time analogue of Hawkes processes is presented, with Poisson-driven increments, long-memory influence kernels expressed via mixture or binomial-thinning constructions, and GINAR(p) links are explored.
Random (element) matrix theory
Wherein the spectra of large symmetric matrices with independent entries are described by Wigner’s semicircle law, eigenvectors and concentration phenomena are examined, and implications for random projections and orthonormal operators are considered
Correlograms
Wherein the correlogram is presented as the mapping of a deterministic L2 signal to its unnormalized autocovariance function, and its scaling, addition, and Wiener–Khintchine links are noted.
Defining dynamics via Gaussian processes
Wherein Gaussian processes are employed to define dynamical laws, and nonparametric transition and observation densities are learned for state‑space models, while links to variational and particle filters are noted.
Representer theorems
Wherein representer theorems are presented as characterizations of minimizers in reproducing-kernel Hilbert spaces, and links to Gaussian processes and spatial covariance modeling are indicated.
Network firewalls, routing etc
In which years of study are needed have basic online safety
Wherein the practical use of host firewalls is described, and iptables/ufw on Linux and PF on macOS are noted as concrete tools to be configured for routing, VPNs and selective service exposure.
Sneakernets
Intermittent connectivity, mesh networks and the Honda protocol
Wherein clandestine data couriers are evoked, and a moped‑borne youth with a backpack of flash drives is described as a high‑capacity conduit for offline, bidirectional internet distribution exemplified by Cuba’s El Paquete.
Biased sampling models
Wherein biased data collection is examined, post-stratification and hierarchical modelling are invoked in survey and psephology contexts to attempt statistical correction, and cases where repair is impossible are noted.
Go bag, Sydney edition
Wherein 72-hour and indefinite crisis plans are outlined for Sydney, bicycle-portable kits are suggested for urban evacuation, and the possibility of euthanasia medications and corpse considerations is examined
Sandboxing apps
Wherein the practice of running untrusted desktop software in constrained OS namespaces is described, and tools such as Firejail, bubblewrap and Flatpak are noted, with audio and escape trade‑offs being discussed.
EZ cross-platform apps
Low-code/ Rapid Application Development, that works across devices
Wherein low-friction methods for enabling citizen data science via mobile phones are examined, and cross-platform frameworks such as Parse and the corporate Protogrid are cited as means for rapid data acquisition.
Installations (in galleries etc)
Wherein procedures for fitting a compact computer into a gallery are described, Raspberry Pi robustness and kiosk‑mode Chromium deployment for unattended exhibits being outlined with practical links and examples
Fourier interpolation
Wherein Fourier interpolation is treated as spectral resampling by DTFT zero-padding and FFT-based differentiation, boundary discontinuities are handled by windowing to avoid Gibbs phenomenon and Nyquist terms are addressed, yielding a minimum-curvature interpolant
Build tools for data science
Wherein a survey is presented of diverse build tools for data science, and their abilities to express DAGs, run containers, and interface with clusters and data‑versioning systems are catalogued.
Topology, applied to problems I know about
Wherein coarse and induced topologies of networks and metric convergence are considered, and connections to persistent homology and applications in adversarial learning are noted.
Game complexity
Wherein the computational difficulty of locating Nash equilibria is surveyed, the PPAD complexity class is invoked, and links to mechanism design, cooperation, and adversarial training are traced.
Making and its discontents
Also repairers, innovators, maintainers, disruptors and sustainers, but not passive consumers
Wherein guidance on 3D printing services in New South Wales is provided and the disposable maintenance of academic projects is noted, exemplified by Leigh Russell (getitmade) being listed for Sydney services.
Wiener theorem
Wherein the special deterministic Wiener theorem is presented and the Fourier transform of the autocorrelogram is shown to be the squared modulus of the signal’s Fourier transform, yielding a power‑spectrum relation
Wacky regression
Wherein a once‑favoured class of almost nonparametric regressions is abandoned and is apportioned into conventional families, with bagging, boosting, neural networks and Gaussian process methods enumerated.
Statistics software
Wherein various statistical and machine‑learning packages are surveyed, including languages (R, Python, Julia, Scala) and tools for streaming, embedded devices, and out‑of‑core learning.
Low impact fashion
Wherein second-hand garments are made distinctive by embroidered patches, eco glitter is employed for cosmetics, and broken ceramics are repaired with kintsugi-style joins, all practices being described.
Ordinary differential equations
Thou, silent form, dost tease us out of thought / As doth eternity
Wherein a pragmatic introductory course by Homer Reid is cited, and extensions to fractional, stochastic, and partial differential equations are indicated as directions for further exposition.
Hollow states
Skim government when fat cats drank the cream
Wherein the mechanisms of governance are described as being hollowed by state capture, exemplified by kleptocratic elites using patronage and opaque loans to secure monopolies such as uranium contracts.
Csound
a less irritating audio programming language
Source: Cabbage audio.
Point processes
Wherein discrete-state random fields with a continuous index are considered, temporal cases being distinguished by past-conditioned intensities allowing explicit log-likelihood integrals for observed event times, with a focus on branching processes.
Inner product spaces
The most highly developed theory of squaring things
Wherein the inner structure of function spaces is set forth and the Riesz representation theorem is exhibited, by which bounded linear functionals are represented as inner products with a fixed element.
Javascript
Wherein a modest compendium of domain-specific JavaScript tricks is presented, with webpack bundling, npm workflow, and assorted developer tools and transpiler notes for web and headless scripting being catalogued in laconic detail.
Dictionaries
Wherein the scarcity of OS‑integratable translation dictionary data files is examined, and practical conversion tools such as pyglossary and xdxf are noted for producing Apple and Kindle‑ready dictionary files.
Normed spaces
Wherein the structure of vector spaces is recalled, norms are introduced to measure size and similarity, and completeness yielding Banach spaces as well as integral kernel operators on L2 are presented.
Machine vision
Wherein optical-flow methods and webcam-driver techniques are surveyed in a compendium of software resources, and practical links to libraries such as OpenCV and ilastik are provided.
Gesture recognition
Wherein gesture recognition for real‑time artistic control is treated, and concrete techniques such as Wekinator workflows, particle‑filter variation following, and the training‑data bottleneck are cited.
The right words
That which I substitute with the long words
Wherein a collection of snippets on the sensation caused by inelegant phrasing is assembled, and examples from Stoppard, Twain, and a statistical tally of gendered verbs in fiction are noted.
Atom
A text editor I seemed to be using
Wherein Atom is described as a JavaScript‑based, open‑source editor created by GitHub, its extensibility via packages and Jupyter (Hydrogen) integration is noted, and its heavy memory use is recorded.
Multiple testing
Wherein the perils of incessant model testing are expounded, and methods for controlling false discoveries—including false discovery rate control and concerns about leaderboard overfitting—are examined
Sparse stochastic processes identification and sampling
Discrete sample representation of sparse continuous stochastic processes
Wherein sampling and estimation for stochastic differential equations driven by Lévy noise are examined, an inference machinery is presented, and Bayesian non‑Gaussian priors for sensing of sparse signals are derived.
Surviving bash
The flagship product of modern unix is certainly better than any other 80s shell
Wherein practical bash survival techniques are set forth, sensible history-search keys are prescribed via .inputrc, and rigorous quoting of variables is enjoined to avert filename-and-whitespace catastrophes.
Pattern formation
Wherein pattern formation is treated via reaction‑diffusion equations and diffusion‑limited aggregation, and evidence from electrical discharge Lichtenberg figures on surfaces is examined.
Field Programmable hardware and ilk
Wherein FPGAs are placed upon the Pareto frontier between cost and speed, and it is recorded that the SPATIAL language is employed to program such devices for somewhat fixed, high‑performance tasks.
10 years of the Living Thing!
Wherein ten years of the Living Thing are noted and its origin is traced to 2009 at the Fenner School of Environment and Society, with URLs migrated from livingthing.org.au to livingthing.org and danmackinlay.name.
Design patterns
Also, life hacks and one-weird-tricks
Wherein a modern catalog by Gordon Brander is contrasted with the aging C2 pattern taxonomy, and a practical, web‑hosted library of software design examples is presented for historical perspective.
Optimisation, combinatorial
Wherein a combinatorial optimization dilemma is presented, and, for certain instances, Google’s OR-Tools are invoked, while occasionally discrete gradient tricks are employed to cause the problem to vanish.
Serious number crunching on Google Cloud
Wherein the use of Google Cloud’s CloudML is examined and its reliance on Python 2.7 and Google Storage APIs is noted, while integration points with TensorFlow, Dataflow, and Datalab are outlined.
Art Python
Wherein Python is deployed to produce audiovisual and graphical works, and practical guidance is imparted on running Python as a web service using tornado and a ProcessPoolExecutor to improve performance.
Sundanese music
Karawitan, Ketuk Tilu, Jaipongan, Gamelan Degung, Death Metal and similar
Wherein Sundanese music is surveyed and its bamboo karinding mouth‑harp is noted as a persistent vernacular instrument, and gamelan Degung, jaipong and trance genres are catalogued with regional tunings.
Submodular functions, maximising
Wherein connections between submodular set functions and convex relaxations are noted, and their role in linking discrete maximisation problems to implicit neural network layers is outlined.
Facebook messages pro-forma response
It doesn’t work but it’s worth trying
Wherein the correspondent is informed that Facebook messages are neglected, and alternate contact is provided by a mobile number reachable via Signal, Telegram, WhatsApp and by email dan@danmackinlay.name
Feature engineering
Wherein automated feature construction from temporal and relational datasets is presented, and entity embeddings and random embeddings are referenced as methods by which categorical and unstructured inputs are represented.
How is foreign filesystem access in macOS awful this week?
Wherein Disk Utility is required to be placed into debug mode to discern volume paths, a mount point outside /Volumes is mandated, and a Linux rescue via VM or USB is employed.
Signal processing
That which you study for 4 years in order to design trippy music visualisers
Wherein the engineering of stochastic time-series inference is presented, with emphasis on linear filters, sampling from continuous to discrete signals, and graph-based signal processing extensions.
Quantum information in physics
Wherein the limits of information storage are examined via the Bekenstein bound, and a ~10^69 bits per square metre ceiling is noted, with black‑hole collapse presented as the enforcing mechanism.
Matlab
A method of charging you licensing fees to use the CPU you already bought
Wherein MATLAB is presented as a numerically focused, commercial programming environment, its proprietary licensing is noted and it is contrasted with Julia as a free, similar-looking alternative.
Rare-event-conditional estimation
Wherein is considered the simulation of quantities conditional on an importance function exceeding a high threshold, and splitting with importance sampling is examined as a remedy for poor Monte Carlo convergence.
Quantum probability
and quantum information, noncommutative probability
Wherein the framework is presented in which classical real-valued probability is subsumed as a special case of complex-valued, non-commutative quantum probability, and measurement is analyzed via logical entropy.
Behavioural economics
Wherein behavioural economics is treated as a modification of classical doctrine, and its transformation into a predictive science via large-scale data and market applications is traced.
Post-selection inference
Adaptive data analysis without cheating
Wherein post-selection inference is considered, and the reusable holdout via differential privacy is introduced as a way to preserve validity across adaptive analyses, with LASSO methods noted.
Model/hyperparameter selection
Wherein the choice among models and hyperparameters is treated as selection of regularisation strength for prediction, and cross-validation, information criteria, and bandit search are described.
Garbled highlights from ICML 2017
Wherein shambolic notes from ICML 2017 in Sydney are recorded, and a tutorial on optimisation is presented with Hessian-eigenvalue criteria for nonconvexity and discussions of annealed objective smoothing.
Gradient descent, constrained
Wherein constrained gradient descent is examined by Lagrange multipliers and Karush–Kuhn–Tucker conditions, saddle points are sought, and primal–dual formulations with L_p norms are considered.
Quantum-probabilistic graphical models
Wherein Reichenbach’s principle is replaced by unitary evolution, as shown by Allen et al., building on the interventionist framework of Costa and Shrapnel, conditional independences are restored and Bayesian inference is enabled for quantum causal models.
Warping and registration problems
Matching up of bumps and wibbles in stretchy things
Wherein affine time-rescalings and smooth nonlinear warps are considered for aligning functions, the computational cost of sinc/spline interpolants is noted, and applications to point processes and DTW are mentioned.
Australian English
Wherein a catalogue of terms such as stickybeak, spruik and rort is presented, and regional dialect variation is mapped across Australia via crowdsourced projects and a Macquarie Dictionary accent map, all being noted.
In Wild Air
Wherein a musician-mathematician is presented, recent work on interactive webcam music is noted, Kasepuhan Cipta Gelar’s 130 varieties of sacred rice are recorded, and bandit algorithms are described.
Chinese language
Wherein the concept of 差不多 is presented as a cultural tendency toward close-enough workmanship, is explained by a Taiwanese flatmate, and is linked to severed feedback loops in regional manufacturing.
Generating functions
Wherein a free textbook by Wilf is referenced, beautiful analytic applications are summoned, and links to asymptotic complex-analysis techniques and graph-theoretic counting notes are indicated.
Lagrangian mechanics
Wherein the calculus of variations is invoked to define an action integral and the principle of least action is employed, and Noether’s theorem is shown to link differentiable symmetries to conserved quantities.
Stochastic processes and fields
Probabilistic structures over index sets and state spaces
Wherein a compendium of stochastic processes is presented, and attention is directed to martingales and Brownian motion as guiding examples, with links to proofs and explanatory essays provided.
Compressed sensing and sampling
Wherein sparsity is exploited to recover signals from few non-local measurements, ℓ1 minimization is presented as a convex surrogate for sparsity, and random measurement matrices are shown to yield suitable restricted isometry constants with high probability.
Marketing psychology
Wherein tactics of persuasion are examined, with privacy intrusions and behavioural-economics nudges considered, a supermarket’s prenatal-targeting case is reported, and randomized trials by Swayable are described.
Linear and least-squares estimation of point processes
Wherein the Berman–Turner device is employed to recast likelihoods as weighted Poisson regressions, empirical K-functions are applied to probe second-order interactions, and practical discretization schemes are described.
Sustainability
Wherein a thousand-year backcast is employed and the problem of costing large-scale economic transitions while minimising human suffering during adaptation to biophysical constraints is dispassionately examined.
Financial stability
When does volatility become a crash?
Wherein 1720 South Sea and Mississippi speculations are surveyed via satirical Dutch engravings from Het Groote Tafereel der Dwaasheid, and the visual record is invoked to trace episodes of market panic and asset bubbles.
Uncertainty principles
Wherein the Fourier relation between conjugate variables is set forth, an entropic lower bound log(e/2) for the sum of position and momentum distributions is established, and equality is exhibited by Gaussians.
Granger causation/Transfer Entropy
Wherein transfer entropy is presented as a KL-divergence between two Markov models for next-step prediction, and its equivalence to Granger causality for linear Gaussian AR models is noted when inferred from data.
Musical metrics and manifolds
Wherein musical spaces, metrics and kernels are considered, Plomp and Levelt’s dissonance curves are invoked to induce nearly circular geometries which are shown to recover the equal‑tempered 12‑tone scale.
How is deep learning on Amazon EC2 awful this week?
Wherein manual CUDA and cuDNN installs are endured, TensorFlow 1.0 is forced to be built from source, ownCloud sync is defeated, and $37.28 plus ~32 hours of work are expended on a single run.
Serious number crunching on Amazon Web Services
Wherein Amazon’s Deep Learning AMIs are examined and outdated or mismatched GPU drivers are revealed, Docker and nvidia-docker deployment paths are surveyed, and Lambda, ECS and EC2 tradeoffs are outlined.
How is Google Cloud ML awful this week?
Wherein the author is detained by Google Cloud ML, is compelled to rewrite laptop code for older Python and local files, is confronted with opaque tooling, and finds promised GPUs unavailable.
Scala
Wherein Scala is presented as a JVM language, and its adoption with Apache Spark for distributed computation and Breeze for numerical statistics is noted, with Darren Wilkinson’s introductions being cited.
Functional equations
Wherein compositional logarithms and exponentials of functions are considered, a method for obtaining functional square roots by an analogue of exp(½ log) is exhibited, and Shannon entropy is derived.
Fractional Brownian motion
Wherein the process is presented as a nonstationary, self-similar generalization of Brownian motion, and its sample-path roughness is governed by a Hurst parameter H in (0,1).
Quasi Monte Carlo
Wherein a deterministic, Monte Carlo–style sampling is described using low-discrepancy sequences such as Sobol nets and Hammersley point sets, and pre-generated point sets are considered.
Metric entropy
Wherein metric sizes of sets are measured by packing, covering and bracketing numbers in metric spaces, and connections to Rademacher and Gaussian complexities are elucidated.
Garbled highlights from NIPS 2016
Wherein the conference is traversed and sessions are catalogued, and Structured Orthogonal Random Features is presented, reducing kernel approximation time from O(d^2) to O(d log d) and speeding computation.
Synestizer
Wherein an instrument is presented that converts images into sound, an online prototype is hosted for public use, source code is published on GitHub, and an invitation to play and contribute is extended to the reader.
Concatenative synthesis
Wherein timbral qualities are transferred from one sound to another using granular audio-mosaic techniques, and synthesis is achieved by concatenating analyzed grains drawn from a corpus.
Javascript reactive programming and streams
Wherein stream-processing approaches in JavaScript are surveyed, and multiple libraries such as RxJS are catalogued, while visual pipeline programs and interoperation via transducers and Fantasy Land are noted.
Genetic algorithms
Wherein the iterative practice of variation and selection is described as a family of nature‑inspired procedures, and resistance to noisy fitness evaluations is noted as a concrete operational feature
Special functions
Wherein a wunderkammer of useful curves is presented, and Bessel, Gamma, Erf and orthogonal polynomials are catalogued with applications to function approximation and probabilistic integrals.
Greatest hits
Wherein percussion is synthesized by machine learning, models are trained on musical corpora and analysis/resynthesis methods are applied to recreate drum and gamelan textures.
The simplex
Wherein a method for producing a uniform point on the n‑simplex is presented, in which n independent Uniform(0,1) draws are ordered and their successive differences are taken as the simplex coordinates.
Maximum likelihood inference
Wherein maximum-likelihood estimators are presented as extremum estimators whose asymptotic optimality is delineated and Fisher information is invoked to quantify the influence of a datum.
Distributed statistical inference
Wherein distributed statistical inference is treated as a task to be carried out by heterogeneous, ad hoc nodes, and message-passing variational inference alongside tools such as Spark and CoCOA are presented.
Stability (in learning)
Wherein the robustness of an estimate to deletion of a single data point is examined as a criterion for generalisation, and the incompatibility of such stability with sparsity (Lasso) is noted.
Fast multipole methods
Wherein it is shown that fast multipole methods are employed as O(N) fast‑summation schemes, and that Gaussian‑like Mercer kernels are approximated by rapid evaluation of field strengths at many targets.
Women in electronic and popular music
Wherein a shortlist of contemporary practitioners in digital audio interaction and algorithm design is compiled to address a noted lack of racial and gender diversity and to provide contacts for current gig bookings.
Kernel density estimators
Wherein kernel density estimators are presented as data distributions convolved with a kernel, and bandwidth is selected by a self-consistency spectral method that yields effective local sample sizes and corrections.
Change of probability measure for a stochastic process
Wherein continuous monotonic changes of measure are considered to render a stochastic process a martingale, and such changes are effected via Girsanov-type transformations with links to reparameterization and neural diffusion.
Surviving macOS server
Wherein the discontinued Server.app is implicated in spawning collabd and Xcode Server processes, and wiki, servermetricsd, and Xcode services are disabled via serverctl and launchctl.
Blind deconvolution
Wherein blind deconvolution is presented as the simultaneous recovery of a signal and its unknown blur, and the case of reconstructing an instrument and its church echo from one recording is treated.
Dynamical systems
Wherein the linear assumption is relaxed and systems are considered whose state is a measure or symbol, randomness and chaos are invoked, and ergodic questions and embedding issues are sketched.
Maximum processes
Processes which can be represented as the maximum value of some underlying process.
Text processing
Wherein document similarity is examined via cosine similarity of vectorized texts, speech tagging is outlined, and finite-state generation is proposed as a basis for string metrics.
Rummaging in string bags
Models for language generation
Wherein bags of words and string metrics are surveyed, and language learning is connected to tensor decomposition and finite‑state generation, with transformer and Hamming‑distance angles being considered.
“Approximate models”
Wherein a frequentist-esque machinery is described, no true models are assumed, approximations are constructed for machine‑learning goals, and certain guarantees are provided under new assumptions
Syntax
Wherein the relations between natural language and computational complexity are examined, and the Chomsky hierarchy and grammatical induction are invoked as concrete frameworks for analysis.
Function approximation and interpolation
Wherein various methods for approximating functions are surveyed and the choice of smoothing and interpolation parameters is considered, with spline smoothing, radial basis functions, warping of curves, cross‑validation, and approximation error measurement being treated.
Function approximation and interpolation
Wherein a scheme for approximating and interpolating functions is presented, and closed-form differentiation and integration are afforded while a profusion of free parameters is required for higher-dimensional cases.
Clustering
Wherein spectral clustering is presented as an entry point, connections to matrix factorisation are sketched, and graph-Laplacian embeddings for network-derived similarities are described
Iterated function systems
Wherein iterated function systems are presented, their relation to L‑systems and fractals is outlined, and the Barnsley method for inferring affine maps from images is described.
Category theory
Wherein an introductory survey of resources and applications is presented, and the relevance of categorical abstractions to programming, formal syntax, and network descriptions is noted.
Deconvolution
Wherein deconvolution of signals is considered, and Wiener-style inversion methods are set forth for high-dimensional, irregularly sampled data with spatially varying kernels, and regularization tradeoffs are examined.
Bahasa Indonesia
Wherein contemporary Indonesian resources are surveyed and an offline dictionary called ID Dict Box is noted to be widely used despite being riddled with advertising, dubious rights claims, and uneven definitions.
Message queues
Wherein asynchronous message-passing is described as a method of concurrency, and sockets carrying atomic messages over transports such as in‑process, inter‑process, TCP, and multicast are noted.
Foundations of mathematics
Wherein an alternative to ZFC is considered: Homotopy Type Theory, featuring the univalence axiom and higher inductive types, is presented as a potential foundation in place of set theory’s choice disputes.
Art LISP
Wherein Lisp dialects are employed in livecoding environments for sound, music, and graphics, temporal recursion and Scheme-derived bricks are described, and compilation-as-service workflows are outlined.
Count time series models
Wherein is considered a taxonomy of discrete-time, discrete-state stochastic models for count series, and renewal-process constructions are shown to generate stationary counts with chosen marginals and tunable autocovariance.
High frequency time series estimation
Wherein estimation from a single high-frequency time series is considered, and asymptotics in sampling density for processes with jumps such as Lévy models are employed to infer parameters including across series.
Hidden variable formalisms in quantum mechanics
Wherein deterministic hidden variables are considered, it is noted that such formalisms are rendered non‑local and that Bohmian determinism is reliant on an infinite‑dimensional configuration space.
Visuals
Wherein an inventory of creative-coding frameworks is presented, naming an audio-capable Praxis, classic Processing, Cinder mesh guides, and JavaScript-oriented graphics tools for producing pretty visuals
Parking sun
Wherein a solo abstract musical project is disclosed, and Sundanese-influenced electronic compositions with controller mappings and machine-listening experiments are catalogued in terse scratch pads.
Pattern machine
Wherein an algorithmic audiovisual apparatus is devised, generative sound and imagery are produced by code, the project’s source is published on GitHub, and a companion blog is maintained.
Statistical mechanics
Wherein the thermodynamics of computation is surveyed alongside traditional models, and connections to statistical inference, game-theoretic dynamics, quantum information, and biological thermodynamics are delineated.
Ensemble Tikoro collaboration
Wherein a sound drama for guttural choir and electronics is presented alongside live projected illustrations, and a childhood at a roofless Sundanese outhouse is depicted as its communal use is shown to fade.
Scarce urban resources
Wherein urban resources are examined, and parking, traffic, licence regimes and public shade are observed to be inefficiently allocated, with shared autonomous vehicles and zoning trade-offs being presented.
How to make crazy experiment on sound design (sic)
Wherein a workshop is presented at Telkom University in Bandung in 2015, and cheap, free software and DIY microphones are exhibited as means for inventive sound design for creative technology students.
Islam
Wherein distinctions among Shi’a, Sunni and movements such as Wahhabi are set out, maps and diagrams are provided, and a Tatar prayer image is used to illustrate regional practice.
Earthquakes
Wherein the mechanics of seismicity are examined through self-exciting statistical models, the amplification of shaking by alluvial sediments is noted, and large-scale tsunami hazard is evoked.
Stream processing and reactive programming
Wherein stream processing and reactive programming are treated as methods for memory-constrained, online transformation of tokenized inputs such as parse trees, and transducer semantics are disambiguated.
Copula functions
Wherein copula functions are presented as instruments by which dependence is encoded via marginal inverse cumulative distributions on the unit cube, and applications to quantitative risk management are noted.
Sigma algebras, probability spaces, measure theory
Wherein the Kolmogorov framework is surveyed and conditional expectation is presented as an analogue of differentiation, while measure-theoretic caveats for financial applications are noted.
Altruism
Wherein the notion of altruism is presented as the residual of animal behaviour unexplained by self-regard, and the scale of its operation—from genes through individuals to societies—is delineated, and altruistic punishment and the design of collectives are noted.
Geometry of fitness landscapes
Wherein the mapping from genotype to phenotype and the effects of path-dependent niche construction are examined, and genome-encoding degeneracies that fold the search space are described.
Normal accidents
Wherein a household of safety is depicted as being undermined by a single kitchen-sink CPU task whose death is implicated in unintended throttle control, and proprietary coding practices are recorded as divergent.
Artificial chemistry
Wherein interacting particles are represented as strings and are treated in distributed agent-based systems, computations akin to chemical reactions are examined for Turing-complete behavior, and biomimetic algorithms are derived.
Sparse regression for inhomogeneous Hawkes processes
My MSc thesis with Professors Didier Sornette and Sara van de Geer
Wherein a novel sampling scheme for an extensive social-media count dataset is employed, and the identification of branching structure under inhomogeneous conditions is formalized and derived via sparse-regression maximum-likelihood methods.
Complexity
Wherein it is argued that no single universal metric of complexity is attainable, and that distinct domain‑specific measures such as algorithmic complexity and thermodynamic depth are relied upon instead.
Mixing and mastering for dummies
Wherein automated mastering services such as LANDR are observed to be proliferating, algorithmic loudness targets are described, the paradox of homogenized records is noted, and Monolake’s technical perspectives are summoned.
Microbial Ecology
Wherein the planetary-scale microbiome is described as being treated like an economy, its rampant horizontal gene transfer and potential climate-feedbacks are analyzed through economic methods.
Red Wine Headache
Wherein sulfite allergy is distinguished from the malady, and prophylactic aspirin ingestion is reported to prevent onset of the red wine headache when taken prior to drinking.
New media art
Wherein new media art is defined by departments younger than fifty years and is presented as a degentrified practice in which glitches, obsolescence, and bricolage are embraced and exhibited.
Algebra I would like to learn
Wherein free groups and Cayley graphs are considered, cycles in random permutations are counted, and the Poisson-Dirichlet law is invoked for large prime factors.
Flocking
Wherein an agent-based model of collective motion is presented, the Boids rules are applied to spatial dynamics, and experiments mixing live animals with simulated agents are described.
Indonesian food
Wherein Indonesian food is presented through a colonial lens and recipes are offered by Rahung Nasution on Koki Kampungan, with dishes such as Bebek Garo Rica and sotong kari hijau being detailed.
Computational complexity
Wherein the interplay of abstract decision problems and physical limitations is examined, and the prospects of simulating entire worlds via computational models and NP‑complete barriers are considered.
Disruptive technology
Wherein the dynamics of collective learning are examined as constrained by human informational limits and geophysical exergy, and South Korea’s chemical-to-automotive industrial trajectory is considered.
Dialects
Wherein the politics and practicality of adopting local speech are examined, and the exigencies of speaking Swiss German in multilingual Switzerland are presented as a concrete test case for social navigation.
Pure
Wherein a recondite tongue is presented as being founded on term rewriting, endowed with pattern matching and symbolic rewriting, and being JIT‑compiled to native code via LLVM for performance.
Research proposal: Grammars of music
Wherein the question of musical grammars is posed, and probabilistic context-free grammars are proposed to be inferred from symbolic corpora such as MIDI and the Million Song Dataset to compare against Markov models.
Time machine 2012
Wherein a perpetually descending Shepard–Risset glissando is overheard in a half‑demolished warehouse, smashed glass and robo‑wars are hosted, and ephemeral performances are presented before eviction.
Dorkbot group show 2012
Wherein a tartan shopping jeep is employed to project a homespun Snake video game across the back wall, biofeedback via an earring and iPad is presented, and heating‑coil glass instruments are displayed.
Macroeconomics
Wherein the study of aggregate economies is surveyed, and the poor performance of dynamic stochastic general equilibrium model forecasts against observed outcomes is documented.
Configuration space of the economy
Wherein the economy is framed as a bounded configuration space at the edge of the atmosphere, and the informational role of technology and its energy constraints are examined as drivers of its trajectories.
Morandini and Blamey
Wherein a post‑suburban shopfront at Camperdown Park is presented, and a solar lampshade wrapping an incandescent bulb is employed to power reclaimed circuit‑board assemblages that are made to emit intermittent squeals.
Dorkbot group show 2011
Wherein at Serial Space a cluster of salvaged electronics is exhibited, a blank monitor wrapped in polarised film and a backpack theremin are presented, and lo‑fi technoscience is enlisted as spectacle.
Econophysics
Wherein the application of statistical‑mechanics models, notably the Ising model and the Minority Game, to economic phenomena is examined, and tensions with established economic methods are outlined.
Honours thesis
Wherein economic models applied to Australian fisheries policy are presented, and fish population dynamics are analyzed alongside policy instruments, empirical calibration, and acknowledged errata.
Feral
Wherein an itinerant soundscape is conjured for a pocket device, and jungle and city field recordings are reshaped into Markovian birdsong and granular textures that are made responsive to passing sounds
Simulation for the social sciences
Wherein are examined the conditions under which complex human decision rules are reduced to low‑dimensional models, with agent‑based models, Herbert Simon’s docility, and model validation for business decisions noted.
The after party for the American Century
Wherein the author is led to Austin to witness SXSW’s brand-saturated spectacle, and a mobile app that algorithmically remixes ambient sound is observed amid steampunk counter-events.
SXSW from the gutter
Wherein the outsider is advised that festival credentials are porous and entry is earned by queuing or happenstance, while oversized microwaved pizza slices are offered as the prevailing local sustenance.
Transmediale 2010
Wherein a journey to Berlin for Transmediale is recounted, contrasts with a boisterous Newcastle festival are drawn, and the muffling effect of double‑glazed windows on local fervour is noted.
How to discover new media art on the web
Wherein the reader is instructed in the use of syndication and social services such as RSS, Twitter, Delicious, festival aggregators and automated recommendation tools, to be kept abreast of ephemeral networked artworks.
Sofa surfing Electrofringe 2009
Wherein the festival’s centre is shifted east to Renew Newcastle, and disused shopfronts are converted into hand-decorated galleries, cafes and performance spaces, lending the event a domestic air.
Mellifera
a show about VR bees I reviewed
Wherein a virtual bee colony is released into a crowded gallery via Second Life–derived CCTV projections, and its simulated ecosystem is presented as vulnerable to invasive pests and algorithmic imbalance.
The discreet charm of the bourgeoisie robot
Wherein a camera-eyed automaton is concealed by a lace-trimmed bustle, static is hissed by gramophone horns, and passersby are engaged in ELIZA-like exchanges that culminate in a visitor draping it with a scarf.
Jakarta Biennale 2009 (Realtime Magazine)
Wherein the city is interrogated by an art biennale, and artworks are installed even in the Grand Indonesia shopping town between Gucci and a Moulin Rouge–themed food hall, commerce being enlisted.
Jakarta Biennale 2009 (Sentap! Magazine)
Wherein a sprawling Jakarta Biennale is traversed, and a public sticker map of Jabodetabek is installed at the National Gallery, where origins and destinations are marked by visitors and giant “Awas Begal!” signs are emblazoned.
Electrofringe will eat itself
Wherein the festival is observed to be eating itself, Arduino workshops are reported as ubiquitous, locative projects are commodified, and handmade aesthetics are reframed into curated collaborative installations.
Dancing machine
Wherein a host of towering automata in bustled Victorian gowns is encountered beneath a glass dome, their mechanical voices are offered as gramophone drones fill the bay and each machine is observed to curtsy inches away.