Membership Inference Attack on CIFAR-100 (test)
1,800TPR@0.1%FPRDeepLeak
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DeepLeakExplanation Method=GradCAM++2026.01 | 1,800 | 90.5 | 98.3 | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=GradCAM++2026.01 | 1,650 | 89.7 | 96.2 | — | |
| DeepLeakExplanation Method=IG2026.01 | 1,642 | 90.3 | 98.6 | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=VarGrad2026.01 | 1,640 | 88.5 | 95.7 | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=SmoothGrad2026.01 | 1,630 | 87.6 | 96.1 | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=LIME2026.01 | 1,630 | 85.2 | 95.9 | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=IG2026.01 | 1,600 | 87.8 | 95.8 | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=GradCAM2026.01 | 1,580 | 88.7 | 96.2 | — | |
| Liu et al. [18] w/ loss traj.Explanation Method=SHAP2026.01 | 1,570 | 86.1 | 96.1 | — | |
| DeepLeakExplanation Method=SHAP2026.01 | 1,180 | 83.93 | 88.4 | — | |
| DeepLeakExplanation Method=LIME2026.01 | 1,050 | 84.25 | 89.1 | — | |
| DeepLeakExplanation Method=SmoothGrad2026.01 | 681 | 79.9 | 82.7 | — | |
| DeepLeakExplanation Method=GradCAM2026.01 | 678 | 82 | 86 | — | |
| DeepLeakExplanation Method=VarGrad2026.01 | 668 | 77.9 | 84.6 | — | |
| Shokri et al. (expl.)Explanation Method=GradCAM2026.01 | 60 | 77.5 | 84.3 | — | |
| Shokri et al. (expl.)Explanation Method=GradCAM++2026.01 | 60 | 76.5 | 84.2 | — | |
| Shokri et al. (expl.)Explanation Method=SmoothGrad2026.01 | 40 | 74.1 | 79.9 | — | |
| Shokri et al. (expl.)Explanation Method=IG2026.01 | 40 | 71.2 | 77.7 | — | |
| Shokri et al. (expl.)Explanation Method=SHAP2026.01 | 40 | 75.1 | 79.8 | — | |
| Shokri et al. (expl.)Explanation Method=LIME2026.01 | 40 | 74.4 | 78.8 | — | |
| Shokri et al. (expl.)Explanation Method=VarGrad2026.01 | 30 | 75.4 | 81.4 | — | |
| MR-BMIANumber of reference models=64, Backbone=DenseNet-1212025.03 | 15.31 | — | — | 45.57 | |
| RMIANumber of reference models=64, Backbone=DenseNet-1212025.03 | 14.5 | — | — | 36.06 | |
| Attack-RNumber of reference models=64, Backbone=DenseNet-1212025.03 | 14.13 | — | — | 42.02 | |
| LiRANumber of reference models=64, Backbone=DenseNet-1212025.03 | 13.19 | — | — | 43.33 |