Image Classification on CIFAR10 (Accuracy, Efficiency, and Calibration Metrics)
92.62AccuracyCosine (Stable)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Cosine (Stable)Backbone=ResNet502026.03 | 92.62 | 92.62 | 40.6776 | 0.7559 | 5 | |
| Cosine (Unst.)Backbone=ResNet502026.03 | 92.34 | 92.34 | 29.3968 | 0.761 | 5 | |
| Bray-Curtis (Norm.)Backbone=ResNet502026.03 | 91.93 | 91.92 | 45.6222 | 0.7883 | 5 | |
| EuclideanBackbone=ResNet502026.03 | 91.85 | 91.85 | 45.8759 | 0.683 | 25.5 | |
| Chebyshev (Std.)Backbone=ResNet502026.03 | 90.5 | 90.5 | 48.5505 | 0.9995 | 1 | |
| BaselineBackbone=ResNet502026.03 | 84.3 | 84.31 | 48.65 | 0.2572 | 50 | |
| Mahalanobis (Chol.)Architecture=CNN, gCO2eq=1.1139, EV=0.4752, PC90%=502026.03 | 66.44 | 66.42 | — | — | — | |
| Bray-Curtis (Norm.)Architecture=CNN, gCO2eq=1.1489, EV=0.8913, PC90%=4.33332026.03 | 65.97 | 65.51 | — | — | — | |
| Minkowski (p=3.0)Architecture=CNN, gCO2eq=1.1598, EV=0.5425, PC90%=502026.03 | 65.89 | 65.93 | — | — | — | |
| Cosine (Stable)Architecture=CNN, gCO2eq=1.1663, EV=0.647, PC90%=18.66672026.03 | 65.84 | 65.66 | — | — | — | |
| EuclideanArchitecture=CNN, gCO2eq=1.1228, EV=0.6582, PC90%=14.33332026.03 | 64.95 | 64.76 | — | — | — | |
| BaselineArchitecture=CNN, gCO2eq=1.12, EV=0.688081, PC90%=9.02026.03 | 62.78 | 62.69 | — | — | — | |
| BaselineBackbone=MLP2026.03 | 53.97 | 53.85 | 0.53 | 0.3465 | 47 | |
| Bray-Curtis (Norm.)Backbone=MLP2026.03 | 52.24 | 52.01 | 0.5264 | 0.967 | 1 | |
| Mahalanobis (Chol.)Backbone=MLP2026.03 | 50.87 | 50.88 | 0.458 | 0.0522 | 50 | |
| Bray-Curtis (Abs.)Backbone=MLP2026.03 | 49.34 | 49.24 | 0.6313 | 0.2434 | 50 | |
| Bray-Curtis (Std.)Backbone=MLP2026.03 | 49.31 | 49.35 | 0.6435 | 0.2906 | 50 | |
| EuclideanBackbone=MLP2026.03 | 48.71 | 48.52 | 0.4303 | 0.4303 | 42.3333 |