Abstract:Membership Inference Attacks (MIAs) aim to estimate whether a specific data point was used in the training of a given model. Existing state-of-the-art attacks typically rely on training multiple reference models to approximate the conditional score distribution for individual data points, which leads to significant computational overhead and limits their practical applicability. In this work, we propose a novel approach -- Bayesian Membership Inference Attack (BMIA), which performs conditional attack through Bayesian sampling. Specifically, we apply Laplace approximation to a single reference model to obtain a posterior over model parameters, enabling direct estimation of the conditional score distribution. Theoretically, we demonstrate that Bayesian sampling reduces intra-model variance, thereby improving attack power. This insight naturally motivates the multi-reference variant that further enhances performance when additional reference models are available. Extensive experiments across image, text, and tabular datasets indicate that our method achieves state-of-the-art performance in both effectiveness and efficiency.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2503.07482 [cs.LG] |
| (or arXiv:2503.07482v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2503.07482 arXiv-issued DOI via DataCite |
From: Zhenlong Liu [view email]
[v1]
Mon, 10 Mar 2025 15:58:43 UTC (1,475 KB)
[v2]
Mon, 25 May 2026 14:35:35 UTC (878 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。