Membership Inference Attack on CIFAR-10 (test)
91Attack AUCDeepLeak
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DeepLeakExplanation Method=SmoothGrad2026.01 | 91 | 1,038 | 78.6 | — | — | |
| DeepLeakExplanation Method=SmoothGrad2026.01 | 91 | 1,038 | 78.6 | — | — | |
| DeepLeakExplanation Method=GradCAM++2026.01 | 86.2 | 670 | 70.3 | — | — | |
| DeepLeakExplanation Method=GradCAM++2026.01 | 86.2 | 670 | 70.3 | — | — | |
| DeepLeakExplanation Method=IG2026.01 | 86.1 | 488 | 79.3 | — | — | |
| DeepLeakExplanation Method=IG2026.01 | 86.1 | 488 | 79.3 | — | — | |
| DeepLeakExplanation Method=SHAP2026.01 | 85.7 | 504 | 71.1 | — | — | |
| DeepLeakExplanation Method=SHAP2026.01 | 85.7 | 504 | 71.1 | — | — | |
| DeepLeakExplanation Method=LIME2026.01 | 81.5 | 1,026 | 75.1 | — | — | |
| DeepLeakExplanation Method=LIME2026.01 | 81.5 | 1,026 | 75.1 | — | — | |
| Liu et al. w/ loss traj.Explanation Method=IG2026.01 | 75.7 | 380 | 65.6 | — | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=IG2026.01 | 75.7 | 380 | 65.6 | — | — | |
| Liu et al. w/ loss traj.Explanation Method=SHAP2026.01 | 75.5 | 390 | 65.2 | — | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=SHAP2026.01 | 75.5 | 390 | 65.2 | — | — | |
| Liu et al. w/ loss traj.Explanation Method=GradCAM2026.01 | 75.1 | 390 | 65.6 | — | — | |
| Liu et al. w/ loss traj.Explanation Method=LIME2026.01 | 75.1 | 400 | 64.4 | — | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=GradCAM2026.01 | 75.1 | 390 | 65.6 | — | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=LIME2026.01 | 75.1 | 400 | 64.4 | — | — | |
| Liu et al. w/ loss traj.Explanation Method=SmoothGrad2026.01 | 75 | 410 | 65.2 | — | — | |
| Liu et al. w/ loss traj.Explanation Method=GradCAM++2026.01 | 75 | 390 | 63.2 | — | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=SmoothGrad2026.01 | 75 | 410 | 65.2 | — | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=GradCAM++2026.01 | 75 | 390 | 63.2 | — | — | |
| Liu et al. w/ loss traj.Explanation Method=VarGrad2026.01 | 74.5 | 390 | 65.6 | — | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=VarGrad2026.01 | 74.5 | 390 | 65.6 | — | — | |
| DeepLeakExplanation Method=VarGrad2026.01 | 73.2 | 678 | 63.5 | — | — | |
| DeepLeakExplanation Method=VarGrad2026.01 | 73.2 | 678 | 63.5 | — | — | |
| Shokri et al. (expl.)Explanation Method=GradCAM++2026.01 | 65.9 | 20 | 62.1 | — | — | |
| Shokri et al. (expl.)Explanation Method=GradCAM++2026.01 | 65.9 | 20 | 62.1 | — | — | |
| Shokri et al. (expl.)Explanation Method=GradCAM2026.01 | 65.4 | 30 | 61.4 | — | — | |
| Shokri et al. (expl.)Explanation Method=GradCAM2026.01 | 65.4 | 30 | 61.4 | — | — | |
| DeepLeakExplanation Method=GradCAM2026.01 | 65.1 | 95 | 63 | — | — | |
| DeepLeakExplanation Method=GradCAM2026.01 | 65.1 | 95 | 63 | — | — | |
| Shokri et al. (expl.)Explanation Method=SmoothGrad2026.01 | 64.2 | 20 | 60.7 | — | — | |
| Shokri et al. (expl.)Explanation Method=SmoothGrad2026.01 | 64.2 | 20 | 60.7 | — | — | |
| Shokri et al. (expl.)Explanation Method=VarGrad2026.01 | 63.8 | 20 | 61.6 | — | — | |
| Shokri et al. (expl.)Explanation Method=VarGrad2026.01 | 63.8 | 20 | 61.6 | — | — | |
| Shokri et al. (expl.)Explanation Method=IG2026.01 | 63 | 30 | 59.4 | — | — | |
| Shokri et al. (expl.)Explanation Method=IG2026.01 | 63 | 30 | 59.4 | — | — | |
| UncodedArchitecture=CNN, Round=302026.07 | 62.02 | — | — | — | — | |
| Shokri et al. (expl.)Explanation Method=SHAP2026.01 | 61.8 | 20 | 60.7 | — | — | |
| Shokri et al. (expl.)Explanation Method=SHAP2026.01 | 61.8 | 20 | 60.7 | — | — | |
| Shokri et al. (expl.)Explanation Method=LIME2026.01 | 61.6 | 10 | 60.4 | — | — | |
| Shokri et al. (expl.)Explanation Method=LIME2026.01 | 61.6 | 10 | 60.4 | — | — | |
| Directional SharpnessMIA Metric=Directional Sharpness, B=12026.06 | 55.6 | — | — | — | 57 | |
| ASAM SharpnessMIA Metric=ASAM Sharpness2026.06 | 53.2 | — | — | — | 53.9 | |
| MKHEArchitecture=CNN, Round=302026.07 | 50.22 | — | — | — | — | |
| GPBACCArchitecture=CNN, Round=302026.07 | 50.18 | — | — | — | — | |
| SVTArchitecture=CNN, Round=302026.07 | 50.12 | — | — | — | — | |
| LossMIA Metric=Loss2026.06 | 49 | — | — | — | 53 | |
| Attack-RNumber of reference models=64, Backbone=ResNet-502025.03 | — | 0.82 | — | 8.23 | — | |
| LiRANumber of reference models=64, Backbone=ResNet-502025.03 | — | 4.08 | — | 11.74 | — | |
| MR-BMIANumber of reference models=64, Backbone=ResNet-502025.03 | — | 4.28 | — | 12.48 | — | |
| RMIANumber of reference models=64, Backbone=ResNet-502025.03 | — | 2.97 | — | 11.15 | — |