Concept Bottleneck Models

Abstract

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.

Edited for clarity from the ICML 2020 version

Cited by in corpus (34)

• WILDS: A Benchmark of in-the-Wild Distribution Shifts
• Logic Explained Networks
• Entropy-based Logic Explanations of Neural Networks
• Model Sketching: Centering Concepts in Early-Stage Machine Learning Model Design
• Now You See Me (CME): Concept-based Model Extraction
• General Multi-label Image Classification with Transformers
• Promises and Pitfalls of Black-Box Concept Learning Models
• Best of both worlds: local and global explanations with human-understandable concepts
• NODE-GAM: Neural Generalized Additive Model for Interpretable Deep Learning
• FUTURE-AI: Guiding Principles and Consensus Recommendations for Trustworthy Artificial Intelligence in Medical Imaging
• Debiasing Concept-based Explanations with Causal Analysis
• Impact of Feedback Type on Explanatory Interactive Learning
• An Integral Projection-based Semantic Autoencoder for Zero-Shot Learning
• Bridging the Gap: Providing Post-Hoc Symbolic Explanations for Sequential Decision-Making Problems with Inscrutable Representations
• Cross-Modal Conceptualization in Bottleneck Models
• Towards Robust Metrics for Concept Representation Evaluation
• Weakly Supervised Multi-task Learning for Concept-based Explainability
• Statistically Significant Concept-based Explanation of Image Classifiers via Model Knockoffs
• PyTorch, Explain! A Python library for Logic Explained Networks
• Towards Human-Interpretable Prototypes for Visual Assessment of Image Classification Models
• Analyzing Effects of Mixed Sample Data Augmentation on Model Interpretability
• Finding and Fixing Spurious Patterns with Explanations
• Toward a Unified Framework for Debugging Concept-based Models
• Composing Neural Learning and Symbolic Reasoning with an Application to Visual Discrimination
• SelfExplain: A Self-Explaining Architecture for Neural Text Classifiers
• Towards Human-Understandable Visual Explanations:Imperceptible High-frequency Cues Can Better Be Removed
• The Definitions of Interpretability and Learning of Interpretable Models
• Learning Interpretable Concept-Based Models with Human Feedback
• Cause and Effect: Hierarchical Concept-based Explanation of Neural Networks
• Failing Conceptually: Concept-Based Explanations of Dataset Shift
• Visual Objectification in Films: Towards a New AI Task for Video Interpretation
• Interpretable Deep Learning for Stock Returns: A Consensus-Bottleneck Asset Pricing Model
• How Faithful are Self-Explainable GNNs?
• A Framework to Learn with Interpretation