Entropy and mutual information in models of deep neural networks

Abstract : We examine a class of stochastic deep learning models with a tractable method to compute information-theoretic quantities. Our contributions are three-fold: (i) We show how entropies and mutual informations can be derived from heuristic statistical physics methods, under the assumption that weight matrices are independent and orthogonally-invariant. (ii) We extend particular cases in which this result is known to be rigorously exact by providing a proof for two-layers networks with Gaussian random weights, using the recently introduced adaptive interpolation method. (iii) We propose an experiment framework with generative models of synthetic datasets, on which we train deep neural networks with a weight constraint designed so that the assumption in (i) is verified during learning. We study the behavior of entropies and mutual informations throughout learning and conclude that, in the proposed setting, the relationship between compression and generalization remains elusive.
Document type :
Preprints, Working Papers, ...
Complete list of metadatas

https://hal-cea.archives-ouvertes.fr/cea-01930228
Contributor : Emmanuelle de Laborderie <>
Submitted on : Wednesday, November 21, 2018 - 4:50:52 PM
Last modification on : Wednesday, May 15, 2019 - 3:41:37 AM
Long-term archiving on : Friday, February 22, 2019 - 3:47:49 PM

File

publi.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : cea-01930228, version 1

Citation

Marylou Gabrié, Andre Manoel, Clément Luneau, Jean Barbier, Nicolas Macris, et al.. Entropy and mutual information in models of deep neural networks. 2018. ⟨cea-01930228⟩

Share

Metrics

Record views

63

Files downloads

127