Multi-objective Neural Architecture Search with Almost No Training

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review

6 Citations (Scopus)

Abstract

In the recent past, neural architecture search (NAS) has attracted increasing attention from both academia and industries. Despite the steady stream of impressive empirical results, most existing NAS algorithms are computationally prohibitive to execute due to the costly iterations of stochastic gradient descent (SGD) training. In this work, we propose an effective alternative, dubbed Random-Weight Evaluation (RWE), to rapidly estimate the performance of network architectures. By just training the last linear classification layer, RWE reduces the computational cost of evaluating an architecture from hours to seconds. When integrated within an evolutionary multi-objective algorithm, RWE obtains a set of efficient architectures with state-of-the-art performance on CIFAR-10 with less than two hours’ searching on a single GPU card. Ablation studies on rank-order correlations and transfer learning experiments to ImageNet have further validated the effectiveness of RWE.

Original languageEnglish
Title of host publicationEvolutionary Multi-Criterion Optimization - 11th International Conference, EMO 2021, Proceedings
EditorsHisao Ishibuchi, Qingfu Zhang, Ran Cheng, Ke Li, Hui Li, Handing Wang, Aimin Zhou
PublisherSpringer Science and Business Media Deutschland GmbH
Pages492-503
Number of pages12
ISBN (Print)9783030720612
DOIs
Publication statusPublished - 2021
Externally publishedYes
Event11th International Conference on Evolutionary Multi-Criterion Optimization, EMO 2021 - Shenzhen, China
Duration: 28 Mar 202131 Mar 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12654 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference11th International Conference on Evolutionary Multi-Criterion Optimization, EMO 2021
Country/TerritoryChina
CityShenzhen
Period28/03/2131/03/21

Keywords

  • Evolutionary algorithms
  • Multi-objective optimization
  • Neural architecture search
  • Performance estimation

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Multi-objective Neural Architecture Search with Almost No Training'. Together they form a unique fingerprint.

Cite this