Geometry-Inspired Top-k Adversarial Perturbations

Nurislam Tursynbek, Aleksandr Petiushko, Ivan Oseledets

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

Abstract

The brittleness of deep image classifiers to small adversarial input perturbations has been extensively studied in the last several years. However, the main objective of existing perturbations is primarily limited to change the correctly predicted Top-1 class by an incorrect one, which does not intend to change the Top-k prediction. In many digital real-world scenarios Top-k prediction is more relevant. In this work, we propose a fast and accurate method of computing Top-k adversarial examples as a simple multi-objective optimization. We demonstrate its efficacy and performance by comparing it to other adversarial example crafting techniques. Moreover, based on this method, we propose Top-k Universal Adversarial Perturbations, image-agnostic tiny perturbations that cause the true class to be absent among the Top-k prediction for the majority of natural images. We experimentally show that our approach outperforms baseline methods and even improves existing techniques of finding Universal Adversarial Perturbations.

Original languageEnglish
Title of host publicationProceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages4059-4068
Number of pages10
ISBN (Electronic)9781665409155
DOIs
Publication statusPublished - 2022
Event22nd IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022 - Waikoloa, United States
Duration: 4 Jan 20228 Jan 2022

Publication series

NameProceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022

Conference

Conference22nd IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022
Country/TerritoryUnited States
CityWaikoloa
Period4/01/228/01/22

Keywords

  • Adversarial Attack and Defense Methods
  • Adversarial Learning
  • Deep Learning

Fingerprint

Dive into the research topics of 'Geometry-Inspired Top-k Adversarial Perturbations'. Together they form a unique fingerprint.

Cite this