Cross-Attack Detection (SSA-CWA vs FOA-Attack) on LLaVA
99PrecisionDense (Ensemble)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Dense (Ensemble)Transfer Setting=SSA-CWA→FOA-Attack2026.05 | 99 | 100 | 99.5 | |
| SAEgis (vision-block0)Transfer Setting=SSA-CWA→FOA-Attack2026.05 | 98 | 98 | 98 | |
| SAEgis (Ensemble)Transfer Setting=SSA-CWA→FOA-Attack2026.05 | 98 | 98 | 98 | |
| PIPTransfer Setting=SSA-CWA→FOA-Attack2026.05 | 97.8 | 89 | 93.1 | |
| SAEgis (projection-mlp2)Transfer Setting=SSA-CWA→FOA-Attack2026.05 | 96.5 | 56 | 70.8 | |
| DenseTransfer Setting=SSA-CWA→FOA-Attack2026.05 | 90.7 | 39 | 54.5 |