Skip to main navigation Skip to search Skip to main content

Anti-bandit Neural Architecture Search for Model Defense

  • Hanlin Chen
  • , Baochang Zhang
  • , Song Xue
  • , Xuan Gong
  • , Hong Liu
  • , Rongrong Ji
  • , David Doermann
  • Beihang University
  • SUNY Buffalo
  • Xiamen University

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

19 Scopus citations

Abstract

Deep convolutional neural networks (DCNNs) have dominated as the best performers in machine learning, but can be challenged by adversarial attacks. In this paper, we defend against adversarial attacks using neural architecture search (NAS) which is based on a comprehensive search of denoising blocks, weight-free operations, Gabor filters and convolutions. The resulting anti-bandit NAS (ABanditNAS) incorporates a new operation evaluation measure and search process based on the lower and upper confidence bounds (LCB and UCB). Unlike the conventional bandit algorithm using UCB for evaluation only, we use UCB to abandon arms for search efficiency and LCB for a fair competition between arms. Extensive experiments demonstrate that ABanditNAS is about twice as fast as the state-of-the-art NAS method, while achieving an 8.73 % improvement over prior arts on CIFAR-10 under PGD-7.

Original languageEnglish
Title of host publicationComputer Vision – ECCV 2020 - 16th European Conference, 2020, Proceedings
EditorsAndrea Vedaldi, Horst Bischof, Thomas Brox, Jan-Michael Frahm
PublisherSpringer Science and Business Media Deutschland GmbH
Pages70-85
Number of pages16
ISBN (Print)9783030586003
DOIs
StatePublished - 2020
Event16th European Conference on Computer Vision, ECCV 2020 - Glasgow, United Kingdom
Duration: Aug 23 2020Aug 28 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12358 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference16th European Conference on Computer Vision, ECCV 2020
Country/TerritoryUnited Kingdom
CityGlasgow
Period08/23/2008/28/20

Keywords

  • Adversarial defense
  • Bandit
  • Neural architecture search (NAS)

Fingerprint

Dive into the research topics of 'Anti-bandit Neural Architecture Search for Model Defense'. Together they form a unique fingerprint.

Cite this