Stable ResNet
Machine Learning
2021-03-19 v2 Machine Learning
Abstract
Deep ResNet architectures have achieved state of the art performance on many tasks. While they solve the problem of gradient vanishing, they might suffer from gradient exploding as the depth becomes large (Yang et al. 2017). Moreover, recent results have shown that ResNet might lose expressivity as the depth goes to infinity (Yang et al. 2017, Hayou et al. 2019). To resolve these issues, we introduce a new class of ResNet architectures, called Stable ResNet, that have the property of stabilizing the gradient while ensuring expressivity in the infinite depth limit.
Cite
@article{arxiv.2010.12859,
title = {Stable ResNet},
author = {Soufiane Hayou and Eugenio Clerico and Bobby He and George Deligiannidis and Arnaud Doucet and Judith Rousseau},
journal= {arXiv preprint arXiv:2010.12859},
year = {2021}
}
Comments
43 pages, 4 figures