No Cover Image

Journal article 723 views 499 downloads

Pruning CNN filters via quantifying the importance of deep visual representations

Ali Alqahtani, Xianghua Xie Orcid Logo, Mark Jones Orcid Logo, Ehab Mohamed Mahmoud Essa Orcid Logo

Computer Vision and Image Understanding, Volume: 208-209, Start page: 103220

Swansea University Authors: Ali Alqahtani, Xianghua Xie Orcid Logo, Mark Jones Orcid Logo, Ehab Mohamed Mahmoud Essa Orcid Logo

  • cviu_swansea.pdf

    PDF | Accepted Manuscript

    ©2021 All rights reserved. All article content, except where otherwise noted, is licensed under a Creative Commons Attribution Non-Commercial No Derivatives License (CC-BY-NC-ND)

    Download (8.25MB)

Abstract

The achievement of convolutional neural networks (CNNs) in a variety of applications is accompanied by a dramatic increase in computational costs and memory requirements. In this paper, we pro- pose a novel framework to measure the importance of individual hidden units by computing a measure of rele...

Full description

Published in: Computer Vision and Image Understanding
ISSN: 1077-3142
Published: Elsevier BV 2021
Online Access: Check full text

URI: https://cronfa.swan.ac.uk/Record/cronfa56831
Tags: Add Tag
No Tags, Be the first to tag this record!
Abstract: The achievement of convolutional neural networks (CNNs) in a variety of applications is accompanied by a dramatic increase in computational costs and memory requirements. In this paper, we pro- pose a novel framework to measure the importance of individual hidden units by computing a measure of relevance to identify the most critical filters and prune them to compress and accelerate CNNs. Unlike exist- ing methods, we introduce the use of the activation of feature maps to detect valuable information and the essential semantic parts to evaluate the importance of feature maps, inspired by novel neural network inter- pretability. A majority voting technique based on the degree of alignment between a semantic concept and individual hidden unit representations is proposed to quantitatively evaluate the importance of feature maps. We also propose a simple yet effective method to estimate new convolution kernels based on the remaining, crucial channels to accomplish effective CNN compression.
Keywords: Deep learning; Convolutional neural networks; Filter pruning; Model compression
College: Faculty of Science and Engineering
Funders: This work is supported by EPSRC, UK EP/N028139/1
Start Page: 103220