Post-hoc Explainability on Nine datasets average
72.2Macro-F1GCE discrete
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GCE discreteRuntime=1.08×2026.05 | 72.2 | 17.6 | 0.5 | 27.7 | 53.3 | |
| Occlusion top-kRuntime=8.70×2026.05 | 66.8 | 5.2 | 2.2 | 13.9 | 40.5 | |
| Integrated gradientsRuntime=3.80×2026.05 | 66.2 | 4.8 | 2.3 | 12.8 | 39.6 | |
| Gradient saliencyRuntime=1.42×2026.05 | 65.5 | 4.2 | 2.5 | 11.8 | 38.6 | |
| CLAM attentionRuntime=1.10×2026.05 | 64.5 | 3.5 | 2.7 | 11.2 | 37.4 | |
| Attention top-kRuntime=1.00×2026.05 | 64 | 3.3 | 2.9 | 10.7 | 36.7 |