Post-hoc Self-explanation of CNNs

2026-03-30

AI Summary

Standard convolutional neural networks lack semantically coherent interpretability due to their internal prototypes’ inability to accurately represent data. This work proposes a unified k-means posterior interpretation framework that replaces the final linear layer of a CNN with a k-means classifier and leverages shallow intermediate feature activations—such as those from the B234 blocks of ResNet34—to generate concept-based explanation maps without requiring gradient computation. By exploiting the spatial consistency inherent in convolutional receptive fields, the method substantially enhances the semantic fidelity of explanations while incurring only a minor degradation in predictive performance, thereby achieving synergistic self-explainability across the encoder, classifier, and feature activation components.

Abstract

Although standard Convolutional Neural Networks (CNNs) can be mathematically reinterpreted as Self-Explainable Models (SEMs), their built-in prototypes do not on their own accurately represent the data. Replacing the final linear layer with a $k$-means-based classifier addresses this limitation without compromising performance. This work introduces a common formalization of $k$-means-based post-hoc explanations for the classifier, the encoder's final output (B4), and combinations of intermediate feature activations. The latter approach leverages the spatial consistency of convolutional receptive fields to generate concept-based explanation maps, which are supported by gradient-free feature attribution maps. Empirical evaluation with a ResNet34 shows that using shallower, less compressed feature activations, such as those from the last three blocks (B234), results in a trade-off between semantic fidelity and a slight reduction in predictive performance.

Problem

• Self-Explainable Models
• Convolutional Neural Networks
• Post-hoc Explanation
• Feature Activations
• Interpretability

Innovation

• post-hoc explanation
• self-explainable models
• k-means classifier
• concept-based explanation
• feature attribution