[2603.28824] SNEAKDOOR: Stealthy Backdoor Attacks against Distribution Matching-based Dataset Condensation

Computer Science > Cryptography and Security arXiv:2603.28824 (cs) [Submitted on 29 Mar 2026] Title:SNEAKDOOR: Stealthy Backdoor Attacks against Distribution Matching-based Dataset Condensation Authors:He Yang, Dongyi Lv, Song Ma, Wei Xi, Jizhong Zhao View a PDF of the paper titled SNEAKDOOR: Stealthy Backdoor Attacks against Distribution Matching-based Dataset Condensation, by He Yang and 4 other authors View PDF HTML (experimental) Abstract:Dataset condensation aims to synthesize compact yet informative datasets that retain the training efficacy of full-scale data, offering substantial gains in efficiency. Recent studies reveal that the condensation process can be vulnerable to backdoor attacks, where malicious triggers are injected into the condensation dataset, manipulating model behavior during inference. While prior approaches have made progress in balancing attack success rate and clean test accuracy, they often fall short in preserving stealthiness, especially in concealing the visual artifacts of condensed data or the perturbations introduced during inference. To address this challenge, we introduce Sneakdoor, which enhances stealthiness without compromising attack effectiveness. Sneakdoor exploits the inherent vulnerability of class decision boundaries and incorporates a generative module that constructs input-aware triggers aligned with local feature geometry, thereby minimizing detectability. This joint design enables the attack to remain imperceptible to both ...