Research map / Label-efficient and robust learning / Recognition and long-tail
Auxiliary Losses for Learning Generalizable Concept-based Models
Ivaxi Sheth, Samira Ebrahimi Kahou
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 concept orthogonal loss (COL) (ablation) | Coop-CBM · CUB | task accuracy: 83.6 -> 84.1 | Improved |
| add COL (ablation) | Coop-CBM · TIL | task accuracy: 53.4 -> 54.2 | Improved |
| add COL (ablation) | Joint CBM · CUB | concept accuracy: 93.2 -> 96.4 | Improved |
| add COL (ablation) | Independent CBM · CUB | concept accuracy: 96.6 -> 97.2 | Mixed |
| add COL (ablation) | Coop-CBM · CUB out-of-domain spurious background | accuracy: 35.4 -> 36.2 | Improved |
| train with hard (clipped) concept values (ablation) | Coop-CBM · CUB | task accuracy: 83.6 -> 82.1 | Hurt |
| clip concept predictions to hard labels at evaluation (ablation) | Coop-CBM · CUB | task accuracy: 83.6 -> 83.0 | Hurt |
| clip concept predictions to hard labels at evaluation (ablation) | Coop-CBM + COL · CUB | task accuracy: 84.1 -> 84.0 | No clear effect |
| raise all loss weights alpha, beta, gamma from 0.01 to 1.0 (ablation) | Coop-CBM + COL · CUB | task accuracy: 84.1 -> 83.0 | Hurt |
| vary COL weight lambda from 0.05 to 10.0 (ablation) | Coop-CBM + COL · CUB | task accuracy: 84.1 -> 83.6 | No clear effect |
| vary COL weight lambda from 0.05 to 1.0 (ablation) | Coop-CBM + COL · TIL | task accuracy: 54.2 -> 53.6 | No clear effect |
| duplicate 10% to 100% of concepts (ablation) | Coop-CBM + COL · CUB | task accuracy: 83.9 -> 83.4 | No clear effect |
Closest papers by meaning
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.90)
- Hybrid Concept Bottleneck Models (CVPR 2025 · similarity 0.89)
- Sample-efficient Learning of Concepts with Theoretical Guarantees: from Data to Concepts without Interventions (NeurIPS 2025 · similarity 0.89)
- A theoretical design of concept sets: improving the predictability of concept bottleneck models (NeurIPS 2024 · similarity 0.89)
- DCBM: Data-Efficient Visual Concept Bottleneck Models (ICML 2025 · similarity 0.89)
- VLG-CBM: Training Concept Bottleneck Models with Vision-Language Guidance (NeurIPS 2024 · similarity 0.89)
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.
