Deep Neural Network Pruning Using Persistent Homology

Satoru Watanabe, Hayato Yamana

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Deep neural networks (DNNs) have improved the performance of artificial intelligence systems in various fields including image analysis, speech recognition, and text classification. However, the consumption of enormous computation resources prevents DNNs from operating on small computers such as edge sensors and handheld devices. Network pruning (NP), which removes parameters from trained DNNs, is one of the prominent methods of reducing the resource consumption of DNNs. In this paper, we propose a novel method of NP, hereafter referred to as PHPM, using persistent homology (PH). PH investigates the inner representation of knowledge in DNNs, and PHPM utilizes the investigation in NP to improve the efficiency of pruning. PHPM prunes DNNs in ascending order of magnitudes of the combinational effects among neurons, which are calculated using the one-dimensional PH, to prevent the deterioration of the accuracy. We compared PHPM with global magnitude pruning method (GMP), which is one of the common baselines to evaluate pruning methods. Evaluation results show that the classification accuracy of DNNs pruned by PHPM outperforms that pruned by GMP.

Original languageEnglish
Title of host publicationProceedings - 2020 IEEE 3rd International Conference on Artificial Intelligence and Knowledge Engineering, AIKE 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages153-156
Number of pages4
ISBN (Electronic)9781728187082
DOIs
Publication statusPublished - 2020 Dec
Event3rd IEEE International Conference on Artificial Intelligence and Knowledge Engineering, AIKE 2020 - Irvine, United States
Duration: 2020 Dec 92020 Dec 11

Publication series

NameProceedings - 2020 IEEE 3rd International Conference on Artificial Intelligence and Knowledge Engineering, AIKE 2020

Conference

Conference3rd IEEE International Conference on Artificial Intelligence and Knowledge Engineering, AIKE 2020
CountryUnited States
CityIrvine
Period20/12/920/12/11

Keywords

  • deep neural network
  • network pruning
  • persistent homology
  • topological data analysis

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Networks and Communications
  • Hardware and Architecture
  • Signal Processing
  • Information Systems and Management

Fingerprint Dive into the research topics of 'Deep Neural Network Pruning Using Persistent Homology'. Together they form a unique fingerprint.

Cite this