notesum.ai
Published at October 21Bayesian Concept Bottleneck Models with LLM Priors
cs.LG
cs.AI
cs.CL
Released Date: October 21, 2024
Authors: Jean Feng1, Avni Kothari2, Luke Zier3, Chandan Singh1, Yan Shuo Tan1
Aff.: 1University of California, San Francisco; 2Microsoft Research; 3National University of Singapore

| Bird | # classes | BC-LLM | One-pass | Boosting | ResNet |
|---|---|---|---|---|---|
| Albatross | 3 | 0.837 (0.771, 0.894) | 0.833 (0.771, 0.890) | 0.775 (0.709, 0.842) | 0.837 (0.767, 0.897) |
| Auklet | 4 | 0.884 (0.830, 0.928) | 0.865 (0.813, 0.914) | 0.733 (0.671, 0.792) | 0.831 (0.767, 0.893) |
| Blackbird | 4 | 0.869 (0.822, 0.912) | 0.719 (0.671, 0.771) | 0.749 (0.689, 0.801) | 0.857 (0.811, 0.900) |
| Bunting | 3 | 0.932 (0.880, 0.972) | 0.629 (0.585, 0.672) | 0.681 (0.593, 0.762) | 0.919 (0.869, 0.962) |
| Flycatcher | 7 | 0.862 (0.835, 0.887) | 0.800 (0.766, 0.833) | 0.784 (0.753, 0.813) | 0.817 (0.781, 0.849) |
| Sparrow | 21 | 0.803 (0.785, 0.819) | 0.723 (0.704, 0.739) | 0.618 (0.597, 0.640) | 0.812 (0.794, 0.830) |
| Tern | 7 | 0.817 (0.784, 0.849) | 0.705 (0.667, 0.740) | 0.609 (0.567, 0.651) | 0.714 (0.674, 0.751) |