Graph Concept Bottleneck Models

TL;DR

GraphCBMs extend Concept Bottleneck Models by incorporating latent concept graphs, capturing concept relationships to improve interpretability and performance in image classification tasks.

Contribution

This paper introduces GraphCBMs, a novel variant of CBMs that models concept relationships with latent graphs, enhancing interpretability and classification accuracy.

Findings

GraphCBMs outperform traditional CBMs in image classification.

Latent concept graphs improve interpretability and intervention effectiveness.

Model demonstrates robustness across various training settings.

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.