Boosting Concept Bottleneck Models with Supervised, Hierarchical Concept Learning

Ao Sun, Yuanyuan Yuan, Pingchuan Ma, Shuai Wang

Keywords: Interpretability, explainability, CBM

Abstract:

Concept Bottleneck Models (CBMs) aim to deliver interpretable and interventionable predictions by bridging features and labels with human-understandable concepts. While recent CBMs show promising potential, they suffer from information leakage, where unintended information beyond the concepts (either in probabilistic or binary-state form) is leaked to the subsequent label prediction. Consequently, distinct classes are falsely classified via indistinguishable concepts, undermining the interpretation and intervention of CBMs.

This paper alleviates the information leakage issue by introducing label supervision in concept prediction and constructing a hierarchical concept set. Accordingly, we propose a new paradigm of CBMs, namely SupCBM, which stands for Structured Understanding of leakage Prevention Concept Bottleneck Model, achieving label prediction via predicted concepts and a deliberately structural-designed intervention matrix. SupCBM focuses on concepts that are mostly relevant to the predicted label and only distinguishes classes when different concepts are presented. Our evaluations show that SupCBM’s label prediction outperforms SOTA CBMs over diverse datasets. Its predicted concepts also exhibit better interpretability. With proper quantification of information leakage in different CBMs, we demonstrate that SupCBM significantly reduces the information leakage.

Primary Area: interpretability and explainable AI

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 5476