EXTRACTING EFFECTIVE SUBNETWORKS WITH GUMBEL-SOFTMAX - Archive ouverte HAL Access content directly
Conference Papers Year : 2022

EXTRACTING EFFECTIVE SUBNETWORKS WITH GUMBEL-SOFTMAX

Mohammed Amine Alaoui
  • Function : Author
Hichem Sahbi
Alice Lebois
  • Function : Author

Abstract

Large and performant neural networks are often overparameterized and can be drastically reduced in size and complexity thanks to pruning. Pruning is a group of methods, which seeks to remove redundant or unnecessary weights or groups of weights in a network. These techniques allow the creation of lightweight networks, which are particularly critical in embedded or mobile applications. In this paper, we devise an alternative pruning method that allows extracting effective subnetworks from larger untrained ones. Our method is stochastic and extracts subnetworks by exploring different topologies which are sampled using Gumbel Softmax. The latter is also used to train probability distributions which measure the relevance of weights in the sampled topologies. The resulting subnetworks are further enhanced using a highly efficient rescaling mechanism that reduces training time and improves performance. Extensive experiments conducted on CIFAR show the outperformance of our subnetwork extraction method against the related work.
Fichier principal
Vignette du fichier
article.pdf (111.22 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03838823 , version 1 (03-11-2022)

Identifiers

Cite

Robin Dupont, Mohammed Amine Alaoui, Hichem Sahbi, Alice Lebois. EXTRACTING EFFECTIVE SUBNETWORKS WITH GUMBEL-SOFTMAX. IEEE International Conference on Image Processing (ICIP), Oct 2022, Bordeaux, France. pp.931-935, ⟨10.1109/ICIP46576.2022.9897718⟩. ⟨hal-03838823⟩
34 View
42 Download

Altmetric

Share

Gmail Facebook X LinkedIn More