Research map / Probabilistic and scientific modeling / Causal discovery
Causal Intervention in Concept Bottleneck Models
Zhiyu Zhu, Jiayu Zhang, Zhibo Jin, Xinyi Wang, Fang Chen, Przemyslaw Biecek, Jianlong Zhou
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 |
|---|---|---|---|
| replace Concept Realignment with training-free Causal Intervention | UCP policy · CBM benchmarks (first 10 steps) | Intervention-AUC@10: 3.90% improvement over CR | Improved |
| use CI instead of other baseline methods | CBM interventions · CBM benchmarks (first 10 steps) | Intervention-AUC@10: 8.59% average improvement | Improved |
| use CI at 15 steps instead of CR | UCP policy · CBM benchmarks (first 15 steps) | Intervention-AUC@15: 9.14% over others, 3.46% over CR | Improved |
| use CI at 20 steps instead of CR | UCP policy · CBM benchmarks (first 20 steps) | Intervention-AUC@20: 8.85% over others, 2.88% over CR | Improved |
| add CI on top of UCP over all steps | UCP · CBM benchmarks (all steps) | Intervention-AUC: 5.07% average gain over mean of competing methods | Improved |
| apply CI instead of CR with Int-CEM on AwA2 | Int-CEM · AwA2 individual | Intervention-AUC: 8424.78 -> 8421.38 | No clear effect |
| raise learning rate from 0.03125 to 0.25 (ablation) | CI with CEM · CUB individual | Intervention-AUC: 10610.90 -> 10661.75 | Improved |
| raise learning rate from 0.03125 to 0.25 (ablation) | CI with CBM · CUB grouped | Intervention-AUC: 2537.13 vs. 2561.61 | Hurt |
| vary learning rate across 0.03125 to 0.25 (ablation) | CI · AwA2/CUB | Intervention-AUC | No clear effect |
| use UCP policy instead of COOP policy (ablation) | CEM · AwA2 individual | Intervention-AUC: 8171.87 -> 8459.34 | Improved |
| use UCP policy instead of COOP policy (ablation) | Int-CEM · CUB individual | Intervention-AUC: 9135.53 -> 11078.61 | Improved |
| use UCP instead of EUDTP policy (ablation) | CBM · CUB grouped | Intervention-AUC: 2506.72 -> 2561.61 | No clear effect |
Closest papers by meaning
- Learning to Intervene on Concept Bottlenecks (ICML 2024 · similarity 0.93)
- Stochastic Concept Bottleneck Models (NeurIPS 2024 · similarity 0.90)
- Causally Reliable Concept Bottleneck Models (NeurIPS 2025 · similarity 0.89)
- Sample-efficient Learning of Concepts with Theoretical Guarantees: from Data to Concepts without Interventions (NeurIPS 2025 · similarity 0.88)
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.88)
- Bayesian Concept Bottleneck Models with LLM Priors (NeurIPS 2025 · similarity 0.88)
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.
