A page where I document what I *don’t* know about kernel approximation.
A page about what I *do* know would be empty.

What I mean is: approximating implicit Mercer kernel feature maps with explicit features; Equivalently, approximating the Gram matrix, which is also related to mixture model inference and clustering.

I’m especially interested in how it might be done using random linear algebra via random embeddings such as random projections.

Also I guess we need to know trade-offs of computational time/space cost versus approximation bounds, so that we can decide when to bother. When is it enough to reduce computational cost merely with support vectors, or to evaluate the kernels efficiently using, e.g. the Fast Gauss Transform and related methods, rather than coming up with alternative kernel bases? (e.g. you don’t want to do the fiddly coding for the Faust Gauss transform)

*Not* the related but distinct (terminological collision)
of approximating functions from convolution kernels.
In kernel approximation we look for approximations, in some sense, to
*the component kernels themselves*.

A good introduction to what is going on is

- Random Fourier Features
- Dustin Tran Random features and kernel functions
- Reflections on Random Kitchen Sinks

## Do you need to even calculate the Gram matrix?

Francis R. Bach and Jordan (2002) does a kernel ICA without calculating the whole Gram matrix (!). Does that fit in here?

## Stationary kernels

I am mostly interested in translation-invariant (i.e. stationary, or homogeneous) kernels, so assume I’m talking about those unless I say otherwise.

Short introduction at scikit-learn kernel approximation page.

See (Drineas and Mahoney 2005; T. Yang et al. 2012, 2012; Vedaldi and Zisserman 2012; Rahimi and Recht 2007, 2009; Alaoui and Mahoney 2014; F. Bach 2015; Francis R. Bach 2013; Solin and Särkkä 2020).

The approximations might be random projection, or random sampling based, e.g. the Nyström method, which is reportedly effectively Monte Carlo integration, although I understand there is an optimisation step too?

I need to work out the difference between random Fourier features, random kitchen sinks, Nyström methods and whatever Smola et al (Schölkopf et al. 1998) call their special case Gaussian approximation. I think Random Fourier features are the same as random kitchen sinks, Rahimi and Recht (2007) and Nyström is different (C. K. Williams and Seeger 2001). When we can exploit (data- or kernel-) structure to do better? (say, (Le, Sarlós, and Smola 2013)) Quasi Monte Carlo can improve on random Monte Carlo? (update: someone already had that idea: (J. Yang et al. 2014)) Or better matrix factorisations?

Vempati et al. (2010):

Recently, Maji and Berg and Vedaldi and Zisserman proposed explicit feature maps to approximate the additive kernels (intersection, \(\chi^2\), etc.) by linear ones, thus enabling the use of fast machine learning technique in a non-linear context. An analogous technique was proposed by Rahimi and Recht (Rahimi and Recht 2007) for the translation invariant RBF kernels. In this paper, we complete the construction and combine the two techniques to obtain explicit feature maps for the generalized RBF kernels.

## Using bases

e.g. Random Fourier Features. See GP basis methods.

## Inner product kernels

See e.g. Koltchinskii and Giné (2000). These have some elegant relations to low-dimensional random projections. How well can we approximate NN kernels?

## Implementations

This is the LIBASKIT set of scalable machine learning and data analysis tools. Currently we provide codes for kernel sums, nearest-neighbors, k-means clustering, kernel regression, and multiclass kernel logistic regression. All codes use OpenMP and MPI for shared memory and distributed memory parallelism.

[…] PNYSTR : (Parallel Nystrom method) Code for kernel summation using the Nystrom method.

## Connections

MDS and Kernel PCA and mixture models are all related in a way I should try to understand when I have a moment.

For the connection with kernel PCA, see Schölkopf et al. (1998) and C. K. I. Williams (2001) for metric multidimensional scaling.

Altschuler et al. (2019) use projection tricks to calculate Sinkhorn divergence.

## References

*arXiv:1411.0306 [Cs, Stat]*, November.

*Advances in Neural Information Processing Systems 32*, 11. Curran Associates, Inc.

*arXiv Preprint arXiv:1502.06800*.

*COLT*, 30:185–209.

*Journal of Machine Learning Research*3 (July): 48.

*Advances in Neural Information Processing Systems*16 (7): 449–56.

*Applied and Computational Harmonic Analysis*28 (2): 131–49.

*Proceedings of The 8th Asian Conference on Machine Learning*, 110–25.

*A Course in Approximation Theory*. American Mathematical Soc.

*Random Matrices: Theory and Applications*02 (04): 1350010.

*arXiv:1703.00864 [Stat]*, March.

*arXiv:1605.09049 [Cs, Stat]*, May.

*arXiv:1702.08159 [Cs]*, February.

*PMLR*.

*Journal of Machine Learning Research*6 (December): 2153–75.

*The Annals of Statistics*38 (1).

*arXiv:1606.05316 [Cs]*, June.

*Artificial Intelligence and Statistics*, 583–91. PMLR.

*Bernoulli*6 (1): 113–67.

*IEEE Transactions on Neural Networks*15 (6): 1517–25.

*Proceedings of the International Conference on Machine Learning*.

*International Conference on Computational Learning Theory*, 154–68. Lecture Notes in Computer Science. Springer.

*arXiv:1608.07597 [Stat]*, August.

*arXiv:1612.06470 [Cs, Stat]*, December.

*Advances in Neural Information Processing Systems*, 1177–84. Curran Associates, Inc.

*2008 46th Annual Allerton Conference on Communication, Control, and Computing*, 555–61.

*Advances in Neural Information Processing Systems*, 1313–20. Curran Associates, Inc.

*Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery*7 (2).

*Mustererkennung 1998*, edited by Paul Levi, Michael Schanz, Rolf-Jürgen Ahlers, and Franz May, 125–32. Informatik Aktuell. Springer Berlin Heidelberg.

*Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond*. MIT Press.

*Artificial Neural Networks — ICANN’97*, edited by Wulfram Gerstner, Alain Germond, Martin Hasler, and Jean-Daniel Nicoud, 583–88. Lecture Notes in Computer Science. Springer Berlin Heidelberg.

*Statistics and Computing*30 (2): 419–46.

*arXiv:2105.08875 [Cs, Math, Stat]*, May.

*arXiv:1702.03877 [Stat]*, February.

*Proceedings of the Thirty-First Conference on Uncertainty in Artificial Intelligence*, 862–71. UAI’15. Arlington, Virginia, USA: AUAI Press.

*Proceedings of the AAAI Conference on Artificial Intelligence*32 (1).

*IEEE Transactions on Pattern Analysis and Machine Intelligence*34 (3): 480–92.

*Procedings of the British Machine Vision Conference 2010*, 2.1–11. Aberystwyth: British Machine Vision Association.

*Advances in Neural Information Processing Systems 13*, edited by T. K. Leen, T. G. Dietterich, and V. Tresp, 46:675–81. MIT Press.

*Advances in Neural Information Processing Systems*, 682–88.

*arXiv:1412.8293 [Cs, Math, Stat]*, December.

*Advances in Neural Information Processing Systems*, 476–84.

*arXiv:1701.02324 [Cs]*.

*Proceedings of the 30th International Conference on Machine Learning (ICML-13)*, 570–78.

*arXiv:1606.07892 [Stat]*, June.

## No comments yet. Why not leave one?