Adversarial Attacks and Defenses for Non-Parametric Two-Sample Tests

• Item type: SIG Technical Reports(1)
• 公開日: 2022-06-20
• タイトル: Adversarial Attacks and Defenses for Non-Parametric Two-Sample Tests
• タイトル / 言語: en
• タイトル: Adversarial Attacks and Defenses for Non-Parametric Two-Sample Tests
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: School of Computing, National University of Singapore
• 著者所属: RIKEN Center for Advanced Intelligence
• 著者所属: DeSI Lab, AAII, University of Technology Sydney; Project (AIP)
• 著者所属: RIKEN Center for Advanced Intelligence/Graduate School of Frontier Sciences, The University of Tokyo
• 著者所属: School of Computing, National University of Singapore
• 著者所属(英): en
• 著者所属(英): School of Computing, National University of Singapore
• 著者所属(英): en
• 著者所属(英): RIKEN Center for Advanced Intelligence
• 著者所属(英): en
• 著者所属(英): DeSI Lab, AAII, University of Technology Sydney; Project (AIP)
• 著者所属(英): en
• 著者所属(英): RIKEN Center for Advanced Intelligence/Graduate School of Frontier Sciences, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): School of Computing, National University of Singapore
• 著者名: Xilie, Xu; Jingfeng, Zhang; Feng, Liu; Masashi, Sugiyama; Mohan, Kankanhalli
• 著者名(英): Xilie, Xu; Jingfeng, Zhang; Feng, Liu; Masashi, Sugiyama; Mohan, Kankanhalli
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Non-parametric two-sample tests (TSTs) that judge whether two sets of samples are drawn from the same distribution, have been widely used in the analysis of critical data. People tend to employ TSTs as trusted basic tools and rarely have any doubt about their reliability. This paper systematically uncovers the failure mode of non-parametric TSTs through adversarial attacks and then proposes corresponding defense strategies. First, we theoretically show that an adversary can upperbound the distributional shift which guarantees the attack’s invisibility. Furthermore, we theoretically find that the adversary can also degrade the lower bound of a TST’s test power, which enables us to iteratively minimize the test criterion in order to search for adversarial pairs. To enable TST-agnostic attacks, we propose an ensemble attack (EA) framework that jointly minimizes the different types of test criteria. Second, to robustify TSTs, we propose a max-min optimization that iteratively generates adversarial pairs to train the deep kernels. Extensive experiments on both simulated and real-world datasets validate the adversarial vulnerabilities of non-parametric TSTs and the effectiveness of our proposed defense.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Non-parametric two-sample tests (TSTs) that judge whether two sets of samples are drawn from the same distribution, have been widely used in the analysis of critical data. People tend to employ TSTs as trusted basic tools and rarely have any doubt about their reliability. This paper systematically uncovers the failure mode of non-parametric TSTs through adversarial attacks and then proposes corresponding defense strategies. First, we theoretically show that an adversary can upperbound the distributional shift which guarantees the attack’s invisibility. Furthermore, we theoretically find that the adversary can also degrade the lower bound of a TST’s test power, which enables us to iteratively minimize the test criterion in order to search for adversarial pairs. To enable TST-agnostic attacks, we propose an ensemble attack (EA) framework that jointly minimizes the different types of test criteria. Second, to robustify TSTs, we propose a max-min optimization that iteratively generates adversarial pairs to train the deep kernels. Extensive experiments on both simulated and real-world datasets validate the adversarial vulnerabilities of non-parametric TSTs and the effectiveness of our proposed defense.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10505667
• 書誌情報: 研究報告数理モデル化と問題解決（MPS）; 巻 2022-MPS-138, 号 5, p. 1-27, 発行日 2022-06-20
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8833
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会