Graph Concept Bottleneck Models

2025-08-19

AI Summary

Existing concept bottleneck models (CBMs) assume conditional independence among concepts, ignoring their intrinsic semantic dependencies—thereby limiting interpretability and robustness. To address this, we propose the Graph-enhanced Concept Bottleneck Model (G-CBM), which explicitly models structural dependencies among latent concepts via a learned concept graph. Methodologically, G-CBM integrates a graph neural network into an end-to-end CBM framework, jointly optimizing concept representations and graph topology to enable precise concept-level interventions and counterfactual reasoning. Evaluated on multiple image classification benchmarks, G-CBM achieves consistent accuracy gains (average +1.8%) over baseline CBMs, yields interpretable hierarchical concept structures, and demonstrates strong generalization across diverse model architectures. Our core contribution is the first formulation of implicit concept relationships as a learnable latent graph structure—bridging the gap between the restrictive independence assumption in conventional CBMs and the rich, interdependent semantics of real-world concepts.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Concept Bottleneck Models (CBMs) provide explicit interpretations for deep neural networks through concepts and allow intervention with concepts to adjust final predictions. Existing CBMs assume concepts are conditionally independent given labels and isolated from each other, ignoring the hidden relationships among concepts. However, the set of concepts in CBMs often has an intrinsic structure where concepts are generally correlated: changing one concept will inherently impact its related concepts. To mitigate this limitation, we propose GraphCBMs: a new variant of CBM that facilitates concept relationships by constructing latent concept graphs, which can be combined with CBMs to enhance model performance while retaining their interpretability. Our experiment results on real-world image classification tasks demonstrate Graph CBMs offer the following benefits: (1) superior in image classification tasks while providing more concept structure information for interpretability; (2) able to utilize latent concept graphs for more effective interventions; and (3) robust in performance across different training and architecture settings.

Problem

• Modeling concept relationships in interpretable neural networks
• Addressing conditional independence assumption in concept bottleneck models
• Enhancing interpretability and intervention via latent concept graphs

Innovation

• GraphCBMs incorporate latent concept graphs
• Enhance interpretability with concept relationships
• Improve intervention effectiveness via structured concepts