UMD: Unsupervised Model Detection for X2X Backdoor Attacks
Jul 24, 2023
Sprecher:innen
Zhen Xiang
Speaker · 0 followers
Zidi Xiong
Speaker · 0 followers
Über
Backdoor (Trojan) attack is a common threat to deep neural networks, where samples from one or more source classes embedded with a backdoor trigger will be misclassified to adversarial target classes. Existing methods for detecting whether a classifier is backdoor attacked can only address attacks with a single adversarial target (e.g., all-to-one attack), while failing against more general X2X attacks with an arbitrary number of source classes each paired with an arbitrary target class. In this paper, we propose UMD, the first Unsupervised Model Detection method that effectively detects X2X backdoor attacks via a joint inference of the adversarial (source, target) class pairs. In particular, we first define a novel transferability statistic to measure and select a subset of putative backdoor class pairs based on our proposed clustering approach. Then, these selected class pairs are jointly assessed based on an aggregation of their reverse-engineered trigger size for detection inference, using a robust and unsupervised anomaly detector we proposed. We conduct comprehensive evaluations on three datasets and show that UMD outperforms SOTA detectors (e.g.) by 50% in model detection accuracy against general A2A attacks on CIFAR-10. We also conduct a series of ablation studies to show the strong detection performance of UMD against X2X attacks under various settings, as well as several adaptive backdoor attacks.Backdoor (Trojan) attack is a common threat to deep neural networks, where samples from one or more source classes embedded with a backdoor trigger will be misclassified to adversarial target classes. Existing methods for detecting whether a classifier is backdoor attacked can only address attacks with a single adversarial target (e.g., all-to-one attack), while failing against more general X2X attacks with an arbitrary number of source classes each paired with an arbitrary target class. In this…
Gefällt euch das Format? Vertraut auf SlidesLive, um euer nächstes Event festzuhalten!
Professionelle Aufzeichnung und Livestreaming – weltweit.
Freigeben
Empfohlene Videos
Präsentationen, deren Thema, Kategorie oder Sprecher:in ähnlich sind
ViT-22B: Scaling Vision Transformers to 22 Billion Parameters
Watch later
Thompson Sampling with Diffusion Generative Prior
Watch later
Advances in Privacy and Federated Learning, with Applications to GBoard
Watch later
A Deep Conjugate Direction Method for Iteratively Solving Linear Systems
Watch later
Ayano Kaneda, …
