Classification on CIFAR-10 (Mean accuracy and Average rank)
97.93Mean AccuracyHybridCBM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HybridCBMBottleneck type=CBMs with soft concepts, Backbone=ViT2026.07 | 97.93 | — | |
| CaBMBottleneck type=Strictly text bottleneck, Protocol=text-only classifier2026.07 | 96.42 | — | |
| PS-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.07 | 89.8 | — | |
| VLG-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.07 | 89.6 | — | |
| DynaTabFeature extraction=ResNet50, Subsampling=11K2026.05 | 88.58 | — | |
| MLPFeature extraction=ResNet50, Subsampling=11K2026.05 | 88.1 | — | |
| LaBoBottleneck type=CBMs with soft concepts, Backbone=CNN2026.07 | 87.5 | — | |
| LF-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.07 | 87.3 | — | |
| TabNetFeature extraction=ResNet50, Subsampling=11K2026.05 | 87.25 | — | |
| LGBMFeature extraction=ResNet50, Subsampling=11K2026.05 | 87.2 | — | |
| LSPINFeature extraction=ResNet50, Subsampling=11K2026.05 | 87.17 | — | |
| LLSPINFeature extraction=ResNet50, Subsampling=11K2026.05 | 86.59 | — | |
| PCBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.07 | 77.7 | — |