Enregistré dans:
Détails bibliographiques
Auteurs principaux: Xie, Xinheng, Yamaguchi, Kureha, Leblanc, Margaux, Malzard, Simon, Chhabra, Varun, Nockles, Victoria, Wu, Yue
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:https://arxiv.org/abs/2409.04982
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866912283529052160
author Xie, Xinheng
Yamaguchi, Kureha
Leblanc, Margaux
Malzard, Simon
Chhabra, Varun
Nockles, Victoria
Wu, Yue
author_facet Xie, Xinheng
Yamaguchi, Kureha
Leblanc, Margaux
Malzard, Simon
Chhabra, Varun
Nockles, Victoria
Wu, Yue
contents The rapid advancement of machine learning technologies raises questions about the security of machine learning models, with respect to both training-time (poisoning) and test-time (evasion, impersonation, and inversion) attacks. Models performing image-related tasks, e.g. detection, and classification, are vulnerable to adversarial attacks that can degrade their performance and produce undesirable outcomes. This paper introduces a novel technique for anomaly detection in images called 2DSig-Detect, which uses a 2D-signature-embedded semi-supervised framework rooted in rough path theory. We demonstrate our method in adversarial settings for training-time and test-time attacks, and benchmark our framework against other state of the art methods. Using 2DSig-Detect for anomaly detection, we show both superior performance and a reduction in the computation time to detect the presence of adversarial perturbations in images.
format Preprint
id arxiv_https___arxiv_org_abs_2409_04982
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle 2DSig-Detect: a semi-supervised framework for anomaly detection on image data using 2D-signatures
Xie, Xinheng
Yamaguchi, Kureha
Leblanc, Margaux
Malzard, Simon
Chhabra, Varun
Nockles, Victoria
Wu, Yue
Computer Vision and Pattern Recognition
Probability
Machine Learning
60L99, 68T10
The rapid advancement of machine learning technologies raises questions about the security of machine learning models, with respect to both training-time (poisoning) and test-time (evasion, impersonation, and inversion) attacks. Models performing image-related tasks, e.g. detection, and classification, are vulnerable to adversarial attacks that can degrade their performance and produce undesirable outcomes. This paper introduces a novel technique for anomaly detection in images called 2DSig-Detect, which uses a 2D-signature-embedded semi-supervised framework rooted in rough path theory. We demonstrate our method in adversarial settings for training-time and test-time attacks, and benchmark our framework against other state of the art methods. Using 2DSig-Detect for anomaly detection, we show both superior performance and a reduction in the computation time to detect the presence of adversarial perturbations in images.
title 2DSig-Detect: a semi-supervised framework for anomaly detection on image data using 2D-signatures
topic Computer Vision and Pattern Recognition
Probability
Machine Learning
60L99, 68T10
url https://arxiv.org/abs/2409.04982