Image Classification on Average Performance (MNIST, C10, C100, SVHN) (test)
51.42AccuracyUA-Backprop
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UA-BackpropAttention mechanism=Attribution map2023.04 | 51.42 | 1.968 | |
| IGAttention mechanism=Attribution map2023.04 | 50.65 | 1.989 | |
| delta-CLUEAttention mechanism=Attribution map2023.04 | 50.43 | 2.033 | |
| Blur IGAttention mechanism=Attribution map2023.04 | 50.29 | 2.095 | |
| GradAttention mechanism=Attribution map2023.04 | 49.71 | 2.06 | |
| FullGradAttention mechanism=Attribution map2023.04 | 49.67 | 2.091 | |
| SmoothGradAttention mechanism=Attribution map2023.04 | 49.41 | 2.1 | |
| CLUEAttention mechanism=Attribution map2023.04 | 49.29 | 2.016 | |
| No attentionAttention mechanism=None2023.04 | 48.77 | 2.096 |