Skip to main navigation Skip to search Skip to main content

A New Training Algorithm Based on Finite-Time Stable Theory for Neural Networks

  • Beihang University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

A new training algorithm based on finite-time stable theory is presented for neural networks in this paper. A new weight dynamic law is designed, two hyper parameters c1 and β are given out, and a new weight update algorithm is established. To verify performance of our new training algorithm, simulations of the classification problem of images in set CIFAR-10 by using VGG16 are considered. Some typical training algorithms such as SGD-M, AdaGrad, Adam and HJB integrated with them are compared to our algorithm. The simulating results show that our algorithm needs fewer epoches to converge and has superior training performance.

Original languageEnglish
Title of host publicationProceedings of 2022 Chinese Intelligent Systems Conference - Volume II
EditorsYingmin Jia, Weicun Zhang, Yongling Fu, Shoujun Zhao
PublisherSpringer Science and Business Media Deutschland GmbH
Pages9-17
Number of pages9
ISBN (Print)9789811962257
DOIs
StatePublished - 2022
Event18th Chinese Intelligent Systems Conference, CISC 2022 - Beijing, China
Duration: 15 Oct 202216 Oct 2022

Publication series

NameLecture Notes in Electrical Engineering
Volume951 LNEE
ISSN (Print)1876-1100
ISSN (Electronic)1876-1119

Conference

Conference18th Chinese Intelligent Systems Conference, CISC 2022
Country/TerritoryChina
CityBeijing
Period15/10/2216/10/22

Keywords

  • Deep learning
  • Finite-time stability
  • Neural networks training algorithm
  • Nonlinear control

Fingerprint

Dive into the research topics of 'A New Training Algorithm Based on Finite-Time Stable Theory for Neural Networks'. Together they form a unique fingerprint.

Cite this