Property Inference (Specific Individuals) on Gender Recognition Downstream 10000 samples (train)
0.96AUC (0.1% Ratio)Manipulated Upstream Model (Zero-Activation Attack)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Manipulated Upstream Model (Zero-Activation Attack)Upstream Task=Face Recognition, Target Property=Specific Individuals2023.03 | 0.96 | 1 | |
| Normal Upstream ModelUpstream Task=Face Recognition, Target Property=Specific Individuals2023.03 | 0.49 | 0.52 |