Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Concept Credible Models for Mitigating Shortcuts
Jiaxuan Wang, Sarah Jabbour, Maggie Makar, Michael Sjoding, Jenna Wiens
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add EYE regularization (lambda>0) to STD(C,X) (ablation) | CCM EYE vs STD(C,X) · CUB clean test | accuracy: 69.7 -> 75.2 | Improved |
| condition on known concepts C in addition to X | STD(X) -> STD(C,X) · CUB clean test | accuracy: 55.8 -> 69.7 | Improved |
| use CCM RES instead of standard training on X | STD(X) · CUB clean test | accuracy: 55.8 -> 76.0 | Improved |
| use CCM EYE instead of standard training on X | STD(X) · CUB biased test | accuracy: 75.7 -> 76.0 | No clear effect |
| use CBM (concepts only) instead of CCM | CBM · CUB biased test | accuracy: 76.0 -> 71.6 | Hurt |
| learn C on biased data (vary T), violating A1 | CCM · CUB clean test | accuracy | Mixed |
| replace dimensions of C with noise, violating A2 | CCM · CUB clean test | accuracy | Mixed |
| CCM EYE on MIMIC with sex shortcut vs baselines | CCM EYE · MIMIC-CXR edema | accuracy under shifted correlation | Improved |
| reverse test correlation relative to training | all methods · MIMIC-CXR edema | accuracy | Hurt |
Closest papers by meaning
- Shortcuts and Identifiability in Concept-based Models from a Neuro-Symbolic Lens (NeurIPS 2025 · similarity 0.87)
- On the Foundations of Shortcut Learning (ICLR 2024 · similarity 0.86)
- Towards Combating Frequency Simplicity-biased Learning for Domain Generalization (NeurIPS 2024 · similarity 0.85)
- Truth over Tricks: Measuring and Mitigating Shortcut Learning in Misinformation Detection (NeurIPS 2025 · similarity 0.84)
- Concept-Based Unsupervised Domain Adaptation (ICML 2025 · similarity 0.84)
- SCISSOR: Mitigating Semantic Bias through Cluster-Aware Siamese Networks for Robust Classification (ICML 2025 · similarity 0.84)
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.
