Research map / Label-efficient and robust learning / Recognition and long-tail
Addressing Leakage in Concept Bottleneck Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace independent concept predictor with autoregressive concept predictor in hard CBM (ablation) | Hard CBM · CUB-2011 | concept accuracy: 68.0±0.2 -> 81.4±0.1 | Helped |
| replace independent concept predictor with autoregressive concept predictor in hard CBM (ablation) | Hard CBM · MIMIC-III EWS | label accuracy: 46.6±0.0 -> 48.5±0.0 | Helped |
| replace independent concept predictor with autoregressive concept predictor in hard CBM (ablation) | Hard CBM · CUB-2011 | label accuracy: 79.5±0.3 -> 81.7±0.2 | Helped |
| use soft concept probabilities with joint training (lambda 0.1) vs hard AR | Hard AR · CUB-2011 | concept accuracy: 81.4±0.1 -> 47.9±0.4 | Hurt |
| soft joint vs hard AR label accuracy | Hard AR · CUB-2011 | label accuracy: 81.7±0.2 vs 82.7±0.2 | No clear effect |
| lower lambda in soft joint CBM (1 to 0.01) (ablation) | Soft Joint CBM · CUB-2011 | concept accuracy: 47.7±0.2 -> 0.1±0.0 | Hurt |
| intervene on concepts in soft sequential/joint CBMs | Soft CBMs · MIMIC-III EWS and CUB-2011 | label accuracy after intervention | Hurt |
| intervene on concepts in hard AR CBM | Hard AR · MIMIC-III EWS and CUB-2011 | label accuracy after intervention | Helped |
| add side-channel latent concepts when concept set is incomplete (ablation) | Hard AR CBM · MIMIC-III EWS and CUB-2011 | label accuracy | Helped |
| train side-channel model on partial concepts, then intervene (ablation) | Hard AR with side-channel · MIMIC-III EWS | intervention accuracy | No clear effect |
Closest papers by meaning
- A theoretical design of concept sets: improving the predictability of concept bottleneck models (NeurIPS 2024 · similarity 0.92)
- Editable Concept Bottleneck Models (ICML 2025 · similarity 0.92)
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.91)
- Stochastic Concept Bottleneck Models (NeurIPS 2024 · similarity 0.91)
- An Analysis of Concept Bottleneck Models: Measuring, Understanding, and Mitigating the Impact of Noisy Annotations (NeurIPS 2025 · similarity 0.91)
- Bayesian Concept Bottleneck Models with LLM Priors (NeurIPS 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.