mlsanityData to 7 Oct 2026

Concept Bottleneck Models

arXiv 2007.04612 ↗First posted 9 Jul 2020cs.LG stat.MLcited by 1,022

Not on the map. Why?

Cited by 48 in Multimodal LLMs / vision-language models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

We seek to learn models that we can interact with using high-level concepts: if the model did not think there was a bone spur in the x-ray, would it still predict severe arthritis? State-of-the-art models today do not typically support the manipulation of concepts like "the existence of bone spurs", as they are trained end-to-end to go directly from raw input (e.g., pixels) to output (e.g., arthritis severity). We revisit the classic idea of first predicting concepts that are provided at training time, and then using these concepts to predict the label. By construction, we can intervene on these concept bottleneck models by editing their predicted concept values and propagating these changes to the final prediction. On x-ray grading and bird identification, concept bottleneck models achieve competitive accuracy with standard end-to-end models, while enabling interpretation in terms of high-level clinical concepts ("bone spurs") or bird attributes ("wing color"). These models also allow for richer human-model interaction: accuracy improves significantly if we can correct model mistakes on concepts at test time.

Key facts. Cited by 1,022 papers in the map (data through 7 Oct 2026). First posted on arXiv 9 Jul 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 46; 2022: 86; 2023: 149; 2024: 205; 2025: 257; 2026: 279.

1,022 papers.

Sort:

By area:

  1. 2022Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Taskcited by 417
  2. 2021Interpretable Machine Learning: Fundamental Principles and 10 Grand Challengescited by 364
  3. 2022Language in a Bottle: Language Model Guided Concept Bottlenecks for Interpretable Image Classificationcited by 310
  4. 2022Post-hoc Concept Bottleneck Modelscited by 289
  5. 2023Label-Free Concept Bottleneck Modelscited by 282
  6. 2023Getting aligned on representational alignmentcited by 142
  7. 2021Promises and Pitfalls of Black-Box Concept Learning Modelscited by 128
  8. 2022Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networkscited by 126
  9. 2023Explainable Artificial Intelligence (XAI) 2.0: A Manifesto of Open Challenges and Interdisciplinary Research Directionscited by 123
  10. 2021Acquisition of Chess Knowledge in AlphaZerocited by 120
  11. 2022Foundations and Trends in Multimodal Machine Learning: Principles, Challenges, and Open Questionscited by 116
  12. 2024Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)cited by 116
38 more
  1. 2021Do Concept Bottleneck Models Learn as Intended?cited by 114
  2. 2022Algorithms to estimate Shapley value feature attributionscited by 114
  3. 2021HIVE: Evaluating the Human Interpretability of Visual Explanationscited by 105
  4. 2023Probabilistic Concept Bottleneck Modelscited by 101
  5. 2021Interpretable Image Classification with Differentiable Prototypes Assignmentcited by 97
  6. 2023Concept-based Explainable Artificial Intelligence: A Surveycited by 97
  7. 2024Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discoverycited by 91
  8. 2021NODE-GAM: Neural Generalized Additive Model for Interpretable Deep Learningcited by 88
  9. 2022Towards Human-centered Explainable AI: A Survey of User Studies for Model Explanationscited by 84
  10. 2021Transparency of Deep Neural Networks for Medical Image Analysis: A Review of Interpretability Methodscited by 83
  11. 2022Concept Bottleneck Model with Additional Unsupervised Conceptscited by 79
  12. 2022Interactive Concept Bottleneck Modelscited by 79
  13. 2022Impossibility Theorems for Feature Attributioncited by 78
  14. 2022GlanceNets: Interpretabile, Leak-proof Concept-based Modelscited by 73
  15. 2023Learning Concise and Descriptive Attributes for Visual Recognitioncited by 73
  16. 2023Discover and Cure: Concept-aware Mitigation of Spurious Correlationcited by 71
  17. 2024VLG-CBM: Training Concept Bottleneck Models with Vision-Language Guidancecited by 65
  18. 2021Editing a classifier by rewriting its prediction rulescited by 64
  19. 2022Global Explainability of GNNs via Logic Combination of Learned Conceptscited by 64
  20. 2022"Help Me Help the AI": Understanding How Explainability Can Support Human-AI Interactioncited by 63
  21. 2023A Closer Look at the Intervention Procedure of Concept Bottleneck Modelscited by 63
  22. 2023Text-To-Concept (and Back) via Cross-Model Alignmentcited by 63
  23. 2023Not All Neuro-Symbolic Concepts Are Created Equal: Analysis and Mitigation of Reasoning Shortcutscited by 63
  24. 2024SelfIE: Self-Interpretation of Large Language Model Embeddingscited by 63
  25. 2025Universal Sparse Autoencoders: Interpretable Cross-Model Concept Alignmentcited by 62
  26. 2022Concept Embedding Models: Beyond the Accuracy-Explainability Trade-Offcited by 61
  27. 2022Post hoc Explanations may be Ineffective for Detecting Unknown Spurious Correlationcited by 59
  28. 2023Diagnosing and Rectifying Vision Models using Languagecited by 58
  29. 2022Concept Activation Regions: A Generalized Framework For Concept-Based Explanationscited by 57
  30. 2022Neural Basis Models for Interpretabilitycited by 56
  31. 2023Interpretable Neural-Symbolic Concept Reasoningcited by 56
  32. 2025Interpreting CLIP with Hierarchical Sparse Autoencoderscited by 54
  33. 2022Concept-level Debugging of Part-Prototype Networkscited by 52
  34. 2022Leveraging Explanations in Interactive Machine Learning: An Overviewcited by 52
  35. 2024Explainable and Interpretable Multimodal Large Language Models: A Comprehensive Surveycited by 52
  36. 2021Entropy-based Logic Explanations of Neural Networkscited by 51
  37. 2024Knowledge Mechanisms in Large Language Models: A Survey and Perspectivecited by 50
  38. 2023Learning Bottleneck Concepts in Image Classificationcited by 49

50 most cited of 1,022 All 1,022 (JSON)