Classification on Ham10000 (test)
92.3AUCG-CBM ResNet-50
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| G-CBM ResNet-50Backbone=ResNet-50, Concept Bottleneck=Interpretable concept-graph2026.07 | 92.3 | — | — | — | 90.9 | |
| CNN MobileNet-V2Backbone=MobileNet-V2, Concept Bottleneck=None2026.07 | 91.2 | — | — | — | 89.2 | |
| G-CBM DenseNet-201Backbone=DenseNet-201, Concept Bottleneck=Interpretable concept-graph2026.07 | 90.7 | — | — | — | 86 | |
| CNN DenseNet-201Backbone=DenseNet-201, Concept Bottleneck=None2026.07 | 90.1 | — | — | — | 87.7 | |
| G-CBM MobileNet-V2Backbone=MobileNet-V2, Concept Bottleneck=Interpretable concept-graph2026.07 | 89.2 | — | — | — | 86.5 | |
| CNN ResNet-50Backbone=ResNet-50, Concept Bottleneck=None2026.07 | 89.1 | — | — | — | 86.5 | |
| MLP-CBM ResNet-50Backbone=ResNet-50, Concept Bottleneck=MLP2026.07 | 85.6 | — | — | — | 79.9 | |
| Linear-CBM ResNet-50Backbone=ResNet-50, Concept Bottleneck=Linear2026.07 | 85 | — | — | — | 76.8 | |
| BEiT-B/16Approach=Weight recycling2026.05 | — | — | — | 18.32 | — | |
| CLIPBackbone=ViT-B/322026.05 | — | — | — | 18.85 | — | |
| CLIPBackbone=ViT-L/14, Resolution=336px2026.05 | — | — | — | 17.72 | — | |
| OxEnsembleEnsemble size=212025.12 | — | 70.7 | 6.7 | — | — | |
| OxonFairEnsemble size=212025.12 | — | 67.9 | 8.2 | — | — | |
| Random2026.05 | — | — | — | 14.28 | — |