Exploring epoch-dependent stochastic residual networks
Computer Vision and Pattern Recognition
2017-04-21 v1
Abstract
The recently proposed stochastic residual networks selectively activate or bypass the layers during training, based on independent stochastic choices, each of which following a probability distribution that is fixed in advance. In this paper we present a first exploration on the use of an epoch-dependent distribution, starting with a higher probability of bypassing deeper layers and then activating them more frequently as training progresses. Preliminary results are mixed, yet they show some potential of adding an epoch-dependent management of distributions, worth of further investigation.
Cite
@article{arxiv.1704.06178,
title = {Exploring epoch-dependent stochastic residual networks},
author = {Fabio Carrara and Andrea Esuli and Fabrizio Falchi and Alejandro Moreo Fernández},
journal= {arXiv preprint arXiv:1704.06178},
year = {2017}
}
Comments
Preliminary report