Prototype-Grounded Concept Models for Verifiable Concept Alignment

TL;DR

This paper introduces Prototype-Grounded Concept Models (PGCMs) that enhance interpretability and verifiability in concept-based models by grounding concepts in visual prototypes, enabling inspection and correction.

Contribution

The paper proposes PGCMs that ground concepts in visual prototypes, allowing verification and targeted intervention, improving interpretability over traditional CBMs.

Findings

PGCMs achieve similar accuracy to state-of-the-art CBMs.

PGCMs substantially improve transparency and interpretability.

PGCMs enable targeted human intervention at the prototype level.

Abstract

Concept Bottleneck Models (CBMs) aim to improve interpretability in Deep Learning by structuring predictions through human-understandable concepts, but they provide no way to verify whether learned concepts align with the human's intended meaning, hurting interpretability. We introduce Prototype-Grounded Concept Models (PGCMs), which ground concepts in learned visual prototypes: image parts that serve as explicit evidence for the concepts. This grounding enables direct inspection of concept semantics and supports targeted human intervention at the prototype level to correct misalignments. Empirically, PGCMs achieve similar predictive performance as state-of-the-art CBMs while substantially improving transparency, interpretability, and intervenability.