Membership Inference on Qwen-14B membership dataset
0.953AUCLogit Lens
Evaluation results
| Method | Links | |
|---|---|---|
| Logit LensFeatures=per-layer (32-48 dims), Protocol=5-fold CV, Sample size=250 samples2026.05 | 0.953 | |
| CRM-LTSFeatures=per-layer (32-48 dims), Protocol=5-fold CV, Sample size=250 samples2026.05 | 0.948 | |
| Attention FlowFeatures=per-layer (32-48 dims), Protocol=5-fold CV, Sample size=250 samples2026.05 | 0.825 | |
| Gradient NormFeatures=per-layer (32-48 dims), Protocol=5-fold CV, Sample size=250 samples2026.05 | 0.5 |