Sparsifying Networks via Subdifferential Inclusion - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2022

Sparsifying Networks via Subdifferential Inclusion

Abstract

Sparsifying deep neural networks is of paramount interest in many areas, especially when those networks have to be implemented on lowmemory devices. In this article, we propose a new formulation of the problem of generating sparse weights for a pre-trained neural network. By leveraging the properties of standard nonlinear activation functions, we show that the problem is equivalent to an approximate subdifferential inclusion problem. The accuracy of the approximation controls the sparsity. We show that the proposed approach is valid for a broad class of activation functions (ReLU, sigmoid, softmax). We propose an iterative optimization algorithm to induce sparsity whose convergence is guaranteed. Because of the algorithm flexibility, the sparsity can be ensured from partial training data in a minibatch manner. To demonstrate the effectiveness of our method, we perform experiments on various networks in different applicative contexts: image classification, speech recognition, natural language processing, and time-series forecasting.
Fichier principal
Vignette du fichier
ICML2021_SIS.pdf (541.1 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-03294543 , version 1 (21-07-2021)

Identifiers

  • HAL Id : hal-03294543 , version 1

Cite

Sagar Verma, Jean-Christophe Pesquet. Sparsifying Networks via Subdifferential Inclusion. ICML 2021 - Thirty-eighth International Conference on Machine Learning, Jul 2021, Virtual, France. ⟨hal-03294543⟩
85 View
43 Download

Share

Gmail Mastodon Facebook X LinkedIn More