A Random Matrix Analysis of Learning with α-Dropout - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2020

A Random Matrix Analysis of Learning with α-Dropout

Résumé

This article studies a one 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 = 0 can drastically improve the generalization performances of the classifier.
Fichier principal
Vignette du fichier
rmt4dropout.pdf (350.06 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02971211 , version 1 (19-10-2020)

Identifiants

  • HAL Id : hal-02971211 , version 1

Citer

Mohamed El Amine Seddik, Romain Couillet, Mohamed Tamaazousti. A Random Matrix Analysis of Learning with α-Dropout. ICML 2020 Workshop Artemiss - 1st Workshop on the Art of Learning with Missing Values, Jul 2020, Virtual, France. ⟨hal-02971211⟩
98 Consultations
65 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More