Research map / Optimization / Implicit bias and generalization
Membership Privacy Risks of Sharpness Aware Minimization
Young In Kim, Andrea Agiollo, Pratiksha Agrawal, Johannes O. Royset, Rajiv Khanna
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 |
|---|---|---|---|
| train with SAM instead of SGD | ResNet-style classifiers · CIFAR-100 | test accuracy: 80.30% -> 81.60% | Improved |
| train with SAM instead of SGD | classifiers · CIFAR-100 confidence MIA | attack accuracy: 77.19% -> 79.10% | Hurt |
| train with SAM instead of SGD | classifiers · EyePacs confidence MIA | attack accuracy: 73.40% -> 77.07% | Hurt |
| train with SAM instead of SGD | classifiers · Purchase-100 | test accuracy: 85.50% -> 85.54% | No clear effect |
| train with SAM instead of SGD | classifiers · CIFAR-100 online LiRA | TPR@0.1% FPR: 27.0% -> 31.0% | Hurt |
| train with SAM instead of SGD | classifiers · CIFAR-10 online LiRA | TPR@0.1% FPR: 8.8% -> 12.5% | Hurt |
| train with SAM instead of SGD | classifiers · Texas-100 online RMIA | TPR@0.1% FPR: 2.9% -> 2.8% | No clear effect |
| train with SAM instead of SGD | classifiers · CIFAR-100 (memorization scores) | memorization score distribution | Improved |
| train with SAM instead of SGD | classifiers · CIFAR-100 test buckets by influence entropy | test accuracy | Mixed |
| model SAM as min-G-norm interpolation with G=I+eta*Sigma | linear interpolating regression · theory | non-member output variance | Improved |
| use SAM geometry instead of Euclidean interpolation | linear interpolating regression · theory | confidence-MIA advantage | Hurt |
| use SAM geometry instead of Euclidean interpolation | linear interpolating regression · theory | likelihood-ratio MIA advantage | Hurt |
Closest papers by meaning
- Assessing Per-Sample Membership Inference Vulnerability without Retraining (NeurIPS 2026 · similarity 0.88)
- Low-Cost High-Power Membership Inference Attacks (ICML 2024 · similarity 0.87)
- Why is SAM Robust to Label Noise? (ICLR 2024 · similarity 0.86)
- Sharpness-Aware Minimization Enhances Feature Quality via Balanced Learning (ICLR 2024 · similarity 0.86)
- Mitigating Privacy Risk in Membership Inference by Convex-Concave Loss (ICML 2024 · similarity 0.86)
- Sharpness-Aware Minimization Efficiently Selects Flatter Minima Late In Training (ICLR 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.
