$DPOT_{L_0}$: Concealing Backdoored model updates in Federated Learning by Data Poisoning with $L_0$-norm-bounded Optimized Triggers
Lists
Only you see this
Abstract
Traditional backdoor attacks in Federated Learning (FL) that rely on fixed trigger patterns and model poisoning exhibit deficient attacking performance against state-of-the-art defenses due to the significant divergence between malicious and benign client model updates. To effectively conceal malicious model updates among benign ones, we propose $DPOT_{L_0}$, a backdoor attack strategy in FL that dynamically constructs a per-round backdoor objective by optimizing an $L_0$-norm-bounded backdoor trigger, making backdoor data have minimal effect on model updates and preserving the global model's main-task performance. We theoretically justify the concealment property of $DPOT_{L_0}$'s model updates in linear models. Our experiments show that $DPOT_{L_0}$, via only a data-poisoning attack, effectively undermines state-of-the-art defenses and outperforms existing backdoor attack techniques on various datasets.
BibTeX
@misc{zhang2025dpot,
author = {Yujie Zhang and Neil Zhenqiang Gong and Michael K. Reiter},
title = {{$DPOT\_\{L\_0\}$: Concealing Backdoored model updates in Federated Learning by Data Poisoning with $L\_0$-norm-bounded Optimized Triggers}},
year = {2025},
url = {https://openreview.net/forum?id=O6OugpQ1Pq},
howpublished = {OpenReview},
note = {Submitted to ICML 2025},
}