Confidence Score based Mini-batch Skipping for CNN Training on Mini-batch Training Environment

Joongho Jo, Jongsun Park

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

Abstract

As the convolutional neural network becomes complex and datasets become huge, a lot of time is spent training the network. In this paper, we propose to mitigate this phenomenon with a mini-batch skipping strategy based on an arithmetic mean of confidence score of images. By skipping the unimportant mini-batch on the training phase, the mini-batch skipping provides saving a lot of time on backpropagation and weight update. We empirically demonstrate the effectiveness of our method with Resnet-18, Resnet-50, and mobilenet-v2 on Cifar-10 and Cifar-100. For Res-net-50, mini-batch skipping gives a 1.39x speedup in training operation without significant accuracy drop.

Original languageEnglish
Title of host publicationProceedings - International SoC Design Conference, ISOCC 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages129-130
Number of pages2
ISBN (Electronic)9781728183312
DOIs
Publication statusPublished - 2020 Oct 21
Event17th International System-on-Chip Design Conference, ISOCC 2020 - Yeosu, Korea, Republic of
Duration: 2020 Oct 212020 Oct 24

Publication series

NameProceedings - International SoC Design Conference, ISOCC 2020

Conference

Conference17th International System-on-Chip Design Conference, ISOCC 2020
CountryKorea, Republic of
CityYeosu
Period20/10/2120/10/24

Keywords

  • Convolutional Neural Network (CNN)
  • mini-batch skipping

ASJC Scopus subject areas

  • Energy Engineering and Power Technology
  • Electrical and Electronic Engineering
  • Instrumentation
  • Artificial Intelligence
  • Hardware and Architecture

Fingerprint Dive into the research topics of 'Confidence Score based Mini-batch Skipping for CNN Training on Mini-batch Training Environment'. Together they form a unique fingerprint.

Cite this