[BRT18] SHADE: Information-Based Regularization for Deep Learning.

Conférence Internationale avec comité de lecture : IEEE International Conference on Image Processing, October 2018, Grece,

Mots clés: Deep Learning, Regularization, Information Bottleneck

Résumé: Regularization is a big issue for training deep neural networks. In this paper, we propose a new information-theory-based regularization scheme named SHADE for SHAnnon DEcay. The originality of the approach is to define a prior based on conditional entropy, which explicitly decouples the learning of invariant representations in the regularizer and the learning of correlations between inputs and labels in the data fitting term. Our second contribution is to derive a stochas- tic version of the regularizer compatible with deep learning, resulting in a tractable training scheme. We empirically validate the efficiency of our approach to improve classification performances compared to standard regularization schemes on several standard architectures.


@inproceedings {
title="{SHADE: Information-Based Regularization for Deep Learning.}",
author=" M. Blot and T. Robert and N. Thome and M. Cord ",
booktitle="{IEEE International Conference on Image Processing}",
address=" Grece",