Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
How does Bayesian Sampling help Membership Inference Attacks?
Zhenlong Liu, Wenyu Jiang, Feng Zhou, Hongxin Wei
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Laplace approximation with MC Dropout (ablation) | BMIA · CIFAR-10 | TPR@1%FPR (%): 9.48 -> 7.94 | Hurt |
| replace Laplace approximation with SWAG (ablation) | BMIA · CIFAR-10 | TPR@1%FPR (%): 9.48 -> 6.61 | Hurt |
| replace Laplace approximation with BatchEnsemble (ablation) | BMIA · CIFAR-10 | TPR@1%FPR (%): 9.48 -> 1.41 | Hurt |
| increase number of posterior samples M (ablation) | BMIA · CIFAR-10/100 | TPR | Improved |
| use KFAC instead of diagonal Hessian factorization (ablation) | BMIA · CIFAR-10/100 | TPR | No clear effect |
| use Bayesian sampling on one reference model instead of RMIA with one reference model | RMIA · CIFAR-100 | TPR@1%FPR (%): 10.08 -> 35.75 | Improved |
| use Bayesian sampling on one reference model instead of QMIA | QMIA · CIFAR-100 | TPR@1%FPR (%): 15.26 -> 35.75 | Improved |
| use BMIA with 1 reference model instead of LiRA with 8 reference models | LiRA (n=8) · CIFAR-100 | TPR@1%FPR (%): 23.20 -> 35.75 | Improved |
| use BMIA with 1 reference model instead of LiRA with 8 reference models | LiRA (n=8) · Purchase-100 | TPR@1%FPR (%): 4.60 -> 4.61 | No clear effect |
| use BMIA with 1 reference model instead of RMIA with 8 reference models | RMIA (n=8) · ImageNet | TPR@0.1%FPR (%): 4.39 -> 4.39 | No clear effect |
| use multi-reference MR-BMIA with 64 reference models | LiRA · CIFAR-100 | TPR@1%FPR (%): 43.33 -> 45.57 | Improved |
| increase reference models for LiRA from n=2 to n=8 | LiRA · CIFAR-100 | TPR@1%FPR (%): 3.67 -> 23.20 | Mixed |
Closest papers by meaning
- Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibration (NeurIPS 2024 · similarity 0.89)
- Practical Bayes-Optimal Membership Inference Attacks (NeurIPS 2025 · similarity 0.89)
- Causal Evaluation of Membership Inference Attacks (NeurIPS 2026 · similarity 0.88)
- Low-Cost High-Power Membership Inference Attacks (ICML 2024 · similarity 0.87)
- Estimating Model-Level Membership Inference Vulnerability Without Reference Models (NeurIPS 2026 · similarity 0.87)
- DF-MIA: A Distribution-Free Membership Inference Attack on Fine-Tuned Large Language Models (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
