Post-hoc Self-explanation of CNNs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Boubekki, Ahcène, Clemmensen, Line H.
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918417474256896
author Boubekki, Ahcène
Clemmensen, Line H.
author_facet Boubekki, Ahcène
Clemmensen, Line H.
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2603_28466
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Post-hoc Self-explanation of CNNs
Boubekki, Ahcène
Clemmensen, Line H.
Computer Vision and Pattern Recognition
Machine Learning
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.
title Post-hoc Self-explanation of CNNs
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2603.28466