Research map / Label-efficient and robust learning / Recognition and long-tail
Meaningfully debugging model mistakes using conceptual counterfactual explanations
Abubakar Abid, Mert Yuksekgonul, James Zou
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 |
|---|---|---|---|
| Use joint multiconcept optimization rather than univariate concept scoring (ablation) | CCE · MetaDataset, 20 spurious-correlation scenarios | Precision@3: 0.91 -> 0.95 | Improved |
| Use joint multiconcept optimization rather than univariate concept scoring (ablation) | CCE · MetaDataset, 20 spurious-correlation scenarios | Median rank: 2.00 -> 1.85 | Improved |
| Add validity constraints that prevent adding concepts already present or removing concepts already absent (ablation) | CCE · Controlled dog-concept and two-confounder examples | Plausibility of explanations | Improved |
| Use a control model without the induced spurious correlation | CCE · MetaDataset, 20 spurious-correlation scenarios | Precision@3: 0.95 -> 0.04 | Hurt |
| Use CCE rather than CoCoX | Spurious-correlation identification · MetaDataset, 20 spurious-correlation scenarios | Precision@3: 0.73 -> 0.95 | Improved |
| Use CCE rather than CSS | Spurious-correlation identification · MetaDataset, 20 spurious-correlation scenarios | Precision@3: 0.003 -> 0.95 | Improved |
| Use CCE rather than random concept selection | Spurious-correlation identification · MetaDataset, 20 spurious-correlation scenarios | Precision@3: 0.02 -> 0.95 | Improved |
| Use a concept library containing the target spurious concept (ablation) | CCE · MetaDataset, target concept omitted from the concept bank | Identification of the target concept | Mixed |
| Include the Lateral View concept in the concept library | CCE · SHC pneumothorax, lateral-view mistakes | Identification of Lateral View as largest negative-score concept: 150/150 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CoCoX — CoCoX: Generating Conceptual and Counterfactual Explanations via Fault-Lines (AAAI 2020) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.85)
- LeapFactual: Reliable Visual Counterfactual Explanation Using Conditional Flow Matching (NeurIPS 2025 · similarity 0.85)
- Model Reconstruction Using Counterfactual Explanations: A Perspective From Polytope Theory (NeurIPS 2024 · similarity 0.84)
- Counterfactual Concept Bottleneck Models (ICLR 2025 · similarity 0.84)
- Promoting Counterfactual Robustness through Diversity (AAAI 2024 · similarity 0.83)
- CounterFlowNet: From Minimal Changes to Meaningful Counterfactual Explanations (NeurIPS 2026 · similarity 0.83)
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.
