FaCT: Faithful Concept Traces for Explaining Neural Network Decisions

TL;DR

This paper introduces FaCT, a method for faithful, model-inherent concept explanations in neural networks, enabling consistent interpretation across layers and classes, with a new evaluation metric.

Contribution

FaCT provides a novel approach for concept explanations that are inherently faithful and shared across classes, along with a new metric for evaluating concept consistency.

Findings

FaCT's concepts are more consistent and interpretable than prior methods.

FaCT maintains competitive ImageNet performance.

The C$^2$-Score effectively evaluates concept-based explanations.

Abstract

Deep networks have shown remarkable performance across a wide range of tasks, yet getting a global concept-level understanding of how they function remains a key challenge. Many post-hoc concept-based approaches have been introduced to understand their workings, yet they are not always faithful to the model. Further, they make restrictive assumptions on the concepts a model learns, such as class-specificity, small spatial extent, or alignment to human expectations. In this work, we put emphasis on the faithfulness of such concept-based explanations and propose a new model with model-inherent mechanistic concept-explanations. Our concepts are shared across classes and, from any layer, their contribution to the logit and their input-visualization can be faithfully traced. We also leverage foundation models to propose a new concept-consistency metric, C$^2$-Score, that can be used to evaluate concept-based methods. We show that, compared to prior work, our concepts are quantitatively more consistent and users find our concepts to be more interpretable, all while retaining competitive ImageNet performance.