Simulability on ML-QE (test)
0.8638Pearson CorrelationGradient L2
Evaluation results
| Method | Links | |
|---|---|---|
| Gradient L2Student training samples=8,4002022.04 | 0.8638 | |
| Attention (SMaT)Student training samples=4,2002022.04 | 0.863 | |
| Attention (SMaT)Student training samples=8,4002022.04 | 0.8561 | |
| Gradient L2Student training samples=4,2002022.04 | 0.8535 | |
| Attention (all layers)Student training samples=8,4002022.04 | 0.8467 | |
| Attention (all layers)Student training samples=4,2002022.04 | 0.8193 | |
| Attention (SMaT)Student training samples=2,1002022.04 | 0.8156 | |
| Attention (all layers)Student training samples=2,1002022.04 | 0.812 | |
| Gradient L2Student training samples=2,1002022.04 | 0.8065 | |
| No ExplainerStudent training samples=8,4002022.04 | 0.7891 | |
| Attention (last layer)Student training samples=8,4002022.04 | 0.7798 | |
| Attention (last layer)Student training samples=4,2002022.04 | 0.772 | |
| No ExplainerStudent training samples=4,2002022.04 | 0.7719 | |
| Attention (last layer)Student training samples=2,1002022.04 | 0.7486 | |
| No ExplainerStudent training samples=2,1002022.04 | 0.7457 | |
| Gradient x InputStudent training samples=8,4002022.04 | 0.7141 | |
| Integrated gradientsStudent training samples=4,2002022.04 | 0.7086 | |
| Integrated gradientsStudent training samples=8,4002022.04 | 0.7036 | |
| Gradient x InputStudent training samples=4,2002022.04 | 0.6922 | |
| Gradient x InputStudent training samples=2,1002022.04 | 0.6846 | |
| Integrated gradientsStudent training samples=2,1002022.04 | 0.6686 |