A Random Matrix Analysis of Learning with Α-Dropout

Abstract

This article studies a single hidden layer neural network with generalized Dropout (α-Dropout), where the dropped out features are replaced with an arbitrary value α. Specifically, under a large dimensional data and network regime, we provide the generalization performances for this network on a binary classification problem. We notably demonstrate that a careful choice of α different from 0 can drastically improve the generalization performances of the classifier.

Cite

Text

Seddik et al. "A Random Matrix Analysis of Learning with Α-Dropout." ICML 2020 Workshops: Artemiss, 2020.

Markdown

[Seddik et al. "A Random Matrix Analysis of Learning with Α-Dropout." ICML 2020 Workshops: Artemiss, 2020.](https://mlanthology.org/icmlw/2020/seddik2020icmlw-random/)

BibTeX

@inproceedings{seddik2020icmlw-random,
  title     = {{A Random Matrix Analysis of Learning with Α-Dropout}},
  author    = {Seddik, Mohamed El Amine and Couillet, Romain and Tamaazousti, Mohamed},
  booktitle = {ICML 2020 Workshops: Artemiss},
  year      = {2020},
  url       = {https://mlanthology.org/icmlw/2020/seddik2020icmlw-random/}
}