Recurrent / convolutional / state-space

State space models

Linear systems

See linear feedback systems and linear filter design. for stuff about FIR vs IIR filters.

Koopman operators

Learning state is pointless! infer directly from observations! See Koopmania.

RNNs

Miller and Hardt (2018)

See RNNs.

Transformers

Stability of learning

Hochreiter et al. (2001); Hochreiter (1998); Lamb et al. (2016);Hardt, Ma, and Recht (2018) etc

Stability of dynamics

Conversion between representations

S4

Interesting package of tools from Christopher Ré’s lab, at the intersection of recurrent networks and . See HazyResearch/state-spaces: Sequence Modeling with Structured State Spaces. I find these aesthetically satisfying, because I spent 2 years of my PhD trying to solve the same problem, and failed. These folks did a better job, so I find it slightly validating that the idea was not stupid. Gu et al. (2021):

Recurrent neural networks (RNNs), temporal convolutions, and neural differential equations (NDEs) are popular families of deep learning models for time-series data, each with unique strengths and tradeoffs in modeling power and computational efficiency. We introduce a simple sequence model inspired by control systems that generalizes these approaches while addressing their shortcomings. The Linear State-Space Layer (LSSL) maps a sequence u↦y by simply simulating a linear continuous-time state-space representation x˙=Ax+Bu,y=Cx+Du. Theoretically, we show that LSSL models are closely related to the three aforementioned families of models and inherit their strengths. For example, they generalize convolutions to continuous-time, explain common RNN heuristics, and share features of NDEs such as time-scale adaptation. We then incorporate and generalize recent theory on continuous-time memorization to introduce a trainable subset of structured matrices A that endow LSSLs with long-range memory. Empirically, stacking LSSL layers into a simple deep neural network obtains state-of-the-art results across time series benchmarks for long dependencies in sequential image classification, real-world healthcare regression tasks, and speech. On a difficult speech classification task with length-16000 sequences, LSSL outperforms prior approaches by 24 accuracy points, and even outperforms baselines that use hand-crafted features on 100x shorter sequences.

Gu, Goel, and Ré (2021):

A central goal of sequence modeling is designing a single principled model that can address sequence data across a range of modalities and tasks, particularly on long-range dependencies. Although conventional models including RNNs, CNNs, and Transformers have specialized variants for capturing long dependencies, they still struggle to scale to very long sequences of 10000 or more steps. A promising recent approach proposed modeling sequences by simulating the fundamental state space model (SSM) \(x'(t) = Ax(t) + Bu(t), y(t) = Cx(t) + Du(t)\), and showed that for appropriate choices of the state matrix \(A\), this system could handle long-range dependencies mathematically and empirically. However, this method has prohibitive computation and memory requirements, rendering it infeasible as a general sequence modeling solution. We propose the Structured State Space sequence model (S4) based on a new parameterization for the SSM, and show that it can be computed much more efficiently than prior approaches while preserving their theoretical strengths. Our technique involves conditioning \(A\) with a low-rank correction, allowing it to be diagonalized stably and reducing the SSM to the well-studied computation of a Cauchy kernel. S4 achieves strong empirical results across a diverse range of established benchmarks, including (i) 91% accuracy on sequential CIFAR-10 with no data augmentation or auxiliary losses, on par with a larger 2-D ResNet, (ii) substantially closing the gap to Transformers on image and language modeling tasks, while performing generation 60× faster (iii) SoTA on every task from the Long Range Arena benchmark, including solving the challenging Path-X task of length 16k that all prior work fails on, while being as efficient as all competitors.

Related? Li et al. (2022)

Interesting parallel to the recursive/non-recursive transformer duality in How the RWKV language models. Question: Can they do the jobs of transformers? Nearly (Vardasbi et al. 2023).

References