mlsanityData to 7 Oct 2026

Post-hoc Concept Bottleneck Models

arXiv 2205.15480 ↗First posted 31 May 2022cs.LG cs.AI stat.MLcited by 289

Not on the map. Why?

Abstract (arXiv)

Concept Bottleneck Models (CBMs) map the inputs onto a set of interpretable concepts (``the bottleneck'') and use the concepts to make predictions. A concept bottleneck enhances interpretability since it can be investigated to understand what concepts the model "sees" in an input and which of these concepts are deemed important. However, CBMs are restrictive in practice as they require dense concept annotations in the training data to learn the bottleneck. Moreover, CBMs often do not match the accuracy of an unrestricted neural network, reducing the incentive to deploy them in practice. In this work, we address these limitations of CBMs by introducing Post-hoc Concept Bottleneck models (PCBMs). We show that we can turn any neural network into a PCBM without sacrificing model performance while still retaining the interpretability benefits. When concept annotations are not available on the training data, we show that PCBM can transfer concepts from other datasets or from natural language descriptions of concepts via multimodal models. A key benefit of PCBM is that it enables users to quickly debug and update the model to reduce spurious correlations and improve generalization to new distributions. PCBM allows for global model edits, which can be more efficient than previous works on local interventions that fix a specific prediction. Through a model-editing user study, we show that editing PCBMs via concept-level feedback can provide significant performance gains without using data from the target domain or model retraining.

Key facts. Cited by 289 papers in the map (data through 7 Oct 2026). Builds on 42 of its 68 references that are in the map. First posted on arXiv 31 May 2022.

Builds on

42 of 68 references.

References · 42

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  3. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  4. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  5. 2014Going Deeper with Convolutionscited by 4,231
  6. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,309
  7. 2016ConceptNet 5.5: An Open Multilingual Graph of General Knowledgecited by 1,065
  8. 2020Concept Bottleneck Modelscited by 1,022
  9. 2016COCO-Stuff: Thing and Stuff Classes in Contextcited by 858
  10. 2017Network Dissection: Quantifying Interpretability of Deep Visual Representationscited by 703
  11. 2020Contrastive Learning of Medical Visual Representations from Paired Images and Textcited by 636
  12. 2021Editing Factual Knowledge in Language Modelscited by 548
30 more
  1. 2021Fast Model Editing at Scalecited by 491
  2. 2019Using Pre-Training Can Improve Model Robustness and Uncertaintycited by 373
  3. 2020Pretrained Transformers Improve Out-of-Distribution Robustnesscited by 293
  4. 2019On Completeness-aware Concept-Based Explanations in Deep Neural Networkscited by 239
  5. 2020Concept Whitening for Interpretable Image Recognitioncited by 225
  6. 2020Modifying Memories in Transformer Modelscited by 222
  7. 2021Evaluating Deep Neural Networks Trained on Clinical Images in Dermatology with the Fitzpatrick 17k Datasetcited by 184
  8. 2020Editable Neural Networkscited by 154
  9. 2020A Patient-Centric Dataset of Images and Metadata for Identifying Melanomas Using Clinical Contextcited by 146
  10. 2019Explaining Classifiers with Causal Concept Effect (CaCE)cited by 135
  11. 2018Net2Vec: Quantifying and Explaining how Concepts are Encoded by Filters in Deep Neural Networkscited by 128
  12. 2021Promises and Pitfalls of Black-Box Concept Learning Modelscited by 128
  13. 2021Explaining in Style: Training a GAN to explain a classifier in StyleSpacecited by 122
  14. 2021Acquisition of Chess Knowledge in AlphaZerocited by 120
  15. 2020CausaLM: Causal Model Explanation Through Counterfactual Language Modelscited by 117
  16. 2021Do Concept Bottleneck Models Learn as Intended?cited by 114
  17. 2020Rewriting a Deep Generative Modelcited by 105
  18. 2020Right for the Right Concept: Revising Neuro-Symbolic Concepts by Interacting with their Explanationscited by 94
  19. 2020Don't Judge an Object by Its Context: Learning to Overcome Contextual Biascited by 91
  20. 2022MetaShift: A Dataset of Datasets for Evaluating Contextual Distribution Shifts and Training Conflictscited by 81
  21. 2021Do Language Models Have Beliefs? Methods for Detecting, Updating, and Visualizing Model Beliefscited by 79
  22. 2021IAIA-BL: A Case-based Interpretable Deep Learning Model for Classification of Mass Lesions in Digital Mammographycited by 72
  23. 2021Editing a classifier by rewriting its prediction rulescited by 64
  24. 2021Meaningfully Debugging Model Mistakes using Conceptual Counterfactual Explanationscited by 53
  25. 2019Pretraining boosts out-of-domain robustness for pose estimationcited by 47
  26. 2020FIND: Human-in-the-Loop Debugging Deep Text Classifierscited by 40
  27. 2020On Interpretability of Deep Learning based Skin Lesion Classifiers using Concept Activation Vectorscited by 39
  28. 2021Provable Repair of Deep Neural Networkscited by 39
  29. 2021DISSECT: Disentangled Simultaneous Explanations via Concept Traversalscited by 37
  30. 2021Partially Interpretable Estimators (PIE): Black-Box-Refined Interpretable Machine Learningcited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 5; 2023: 44; 2024: 83; 2025: 75; 2026: 82.

289 papers.

Sort:

By area:

  1. 2023Explainability for Large Language Models: A Surveycited by 331
  2. 2022Language in a Bottle: Language Model Guided Concept Bottlenecks for Interpretable Image Classificationcited by 310
  3. 2023Label-Free Concept Bottleneck Modelscited by 282
  4. 2023Interpreting CLIP's Image Representation via Text-Based Decompositioncited by 189
  5. 2022Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networkscited by 126
  6. 2024Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)cited by 116
  7. 2023Probabilistic Concept Bottleneck Modelscited by 101
  8. 2023Concept-based Explainable Artificial Intelligence: A Surveycited by 97
  9. 2024Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discoverycited by 91
  10. 2023Learning Concise and Descriptive Attributes for Visual Recognitioncited by 73
  11. 2023Discover and Cure: Concept-aware Mitigation of Spurious Correlationcited by 71
  12. 2024VLG-CBM: Training Concept Bottleneck Models with Vision-Language Guidancecited by 65
38 more
  1. 2023A Closer Look at the Intervention Procedure of Concept Bottleneck Modelscited by 63
  2. 2023Text-To-Concept (and Back) via Cross-Model Alignmentcited by 63
  3. 2024Explainable and Interpretable Multimodal Large Language Models: A Comprehensive Surveycited by 52
  4. 2024Stochastic Concept Bottleneck Modelscited by 49
  5. 2023Codebook Features: Sparse and Discrete Interpretability for Neural Networkscited by 45
  6. 2023SkinCon: A skin disease dataset densely annotated by domain experts for fine-grained model debugging and analysiscited by 44
  7. 2024Incremental Residual Concept Bottleneck Modelscited by 43
  8. 2023Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistencycited by 42
  9. 2023Learning to Receive Help: Intervention-Aware Concept Embedding Modelscited by 41
  10. 2024Concept Bottleneck Large Language Modelscited by 41
  11. 2023Robust and Interpretable Medical Image Classifiers via Concept Bottleneck Modelscited by 40
  12. 2023Vamos: Versatile Action Models for Video Understandingcited by 40
  13. 2023Human Uncertainty in Concept-Based AI Systemscited by 39
  14. 2022Augmenting Interpretable Models with LLMs during Trainingcited by 36
  15. 2023Auxiliary Losses for Learning Generalizable Concept-based Modelscited by 35
  16. 2024Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretationscited by 33
  17. 2023Coarse-to-Fine Concept Bottleneck Modelscited by 31
  18. 2023Learning to Intervene on Concept Bottleneckscited by 29
  19. 2024Beyond Concept Bottleneck Models: How to Make Black Boxes Intervenable?cited by 28
  20. 2025Language Guided Concept Bottleneck Models for Interpretable Continual Learningcited by 28
  21. 2023Towards Trustable Skin Cancer Diagnosis via Rewriting Model's Decisioncited by 27
  22. 2024Decomposing Disease Descriptions for Enhanced Pathology Detection: A Multi-Aspect Vision-Language Pre-training Frameworkcited by 27
  23. 2023SI-MIL: Taming Deep MIL for Self-Interpretability in Gigapixel Histopathologycited by 25
  24. 2025Interpretable and Testable Vision Features via Sparse Autoencoderscited by 24
  25. 2023Sparse Linear Concept Discovery Modelscited by 22
  26. 2023Interpretability is in the Mind of the Beholder: A Causal Framework for Human-interpretable Representation Learningcited by 22
  27. 2024Editable Concept Bottleneck Modelscited by 22
  28. 2024Semi-supervised Concept Bottleneck Modelscited by 21
  29. 2024Explain via Any Concept: Concept Bottleneck Model with Open Vocabulary Conceptscited by 20
  30. 2023Theory of Mind as Intrinsic Motivation for Multi-Agent Reinforcement Learningcited by 19
  31. 2024Self-eXplainable AI for Medical Image Analysis: A Survey and New Outlookscited by 18
  32. 2025A Comprehensive Survey on Self-Interpretable Neural Networkscited by 18
  33. 2024Eliminating Information Leakage in Hard Concept Bottleneck Models with Supervised, Hierarchical Concept Learningcited by 17
  34. 2024CoSy: Evaluating Textual Explanations of Neuronscited by 17
  35. 2024Learning Discrete Concepts in Latent Hierarchical Modelscited by 17
  36. 2024Concept Bottleneck Models Without Predefined Conceptscited by 17
  37. 2024Integrating Clinical Knowledge into Concept Bottleneck Modelscited by 17
  38. 2024Concept-Based Explanations in Computer Vision: Where Are We and Where Could We Go?cited by 17

50 most cited of 289 All 289 (JSON)