## highlighted by JP Bouchaud

## [Statistical Criticality arises in Most Informative Representations](https://arxiv.org/pdf/1808.00249.pdf)

- [Why does deep and cheap learning work so well?](https://arxiv.org/pdf/1608.08225.pdf)
Henry W. Lin, Max Tegmark, and David Rolnick

> We explore how properties frequently encountered in physics such as symmetry, locality, compositionality, and
> polynomial log-probability translate into exceptionally simple neural networks

## [Random Matrix Theory Proves that Deep Learning Representations of GAN-data Behave as Gaussian Mixtures](http://proceedings.mlr.press/v119/seddik20a.html)