Wang, H orcid.org/0000-0002-2281-5679, Diao, Y, Tan, Z et al. (1 more author) (2023) Defending Black-box Skeleton-based Human Activity Classifiers. In: Proceedings of the AAAI Conference on Artificial Intelligence. The 37th AAAI conference on Aritificial Intelligence, 07-14 Feb 2023, Washington DC, USA. AAAI , Washington, DC , pp. 2546-2554. ISBN 978-1-57735-880-0
Abstract
Skeletal motions have been heavily relied upon for human activity recognition (HAR). Recently, a universal vulnerability of skeleton-based HAR has been identified across a variety of classifiers and data, calling for mitigation. To this end, we propose the first black-box defense method for skeleton-based HAR to our best knowledge. Our method is featured by full Bayesian treatments of the clean data, the adversaries and the classifier, leading to (1) a new Bayesian Energy-based formulation of robust discriminative classifiers, (2) a new adversary sampling scheme based on natural motion manifolds, and (3) a new post-train Bayesian strategy for black-box defense. We name our framework Bayesian Energy-based Adversarial Training or BEAT. BEAT is straightforward but elegant, which turns vulnerable black-box classifiers into robust ones without sacrificing accuracy. It demonstrates surprising and universal effectiveness across a wide range of skeletal HAR classifiers and datasets, under various attacks. Appendix and code are available.
Metadata
Item Type: | Proceedings Paper |
---|---|
Authors/Creators: |
|
Keywords: | CV: Adversarial Attacks & Robustness, CV: Motion & Tracking |
Dates: |
|
Institution: | The University of Leeds |
Academic Units: | The University of Leeds > Faculty of Engineering & Physical Sciences (Leeds) > School of Computing (Leeds) |
Funding Information: | Funder Grant number EU - European Union 899739 |
Depositing User: | Symplectic Publications |
Date Deposited: | 14 Dec 2022 15:31 |
Last Modified: | 20 Feb 2024 09:28 |
Published Version: | https://ojs.aaai.org/index.php/AAAI/article/view/2... |
Status: | Published |
Publisher: | AAAI |
Identification Number: | 10.1609/aaai.v37i2.25352 |
Open Archives Initiative ID (OAI ID): | oai:eprints.whiterose.ac.uk:193975 |