English

REVE: Regularizing Deep Learning with Variational Entropy Bound

Machine Learning 2019-10-16 v1 Machine Learning

Abstract

Studies on generalization performance of machine learning algorithms under the scope of information theory suggest that compressed representations can guarantee good generalization, inspiring many compression-based regularization methods. In this paper, we introduce REVE, a new regularization scheme. Noting that compressing the representation can be sub-optimal, our first contribution is to identify a variable that is directly responsible for the final prediction. Our method aims at compressing the class conditioned entropy of this latter variable. Second, we introduce a variational upper bound on this conditional entropy term. Finally, we propose a scheme to instantiate a tractable loss that is integrated within the training procedure of the neural network and demonstrate its efficiency on different neural networks and datasets.

Keywords

Cite

@article{arxiv.1910.06816,
  title  = {REVE: Regularizing Deep Learning with Variational Entropy Bound},
  author = {Antoine Saporta and Yifu Chen and Michael Blot and Matthieu Cord},
  journal= {arXiv preprint arXiv:1910.06816},
  year   = {2019}
}

Comments

Published in 2019 IEEE International Conference on Image Processing (ICIP)

R2 v1 2026-06-23T11:44:20.437Z