Linking in Style: Understanding learned features in deep learning models

TL;DR

This paper introduces a novel, efficient method to visualize and analyze learned features in CNNs by linking them to a generative model's latent space, enabling better interpretability and understanding of model representations.

Contribution

We propose a linking network that maps CNN features to StyleGAN-XL's latent space, providing a human-friendly visualization and systematic analysis of learned features in deep models.

Findings

Semantic order is consistent between CNN features and StyleGAN-XL latent space.

The linking network is computationally inexpensive and decoupled from training the GAN and classifier.

Our pipeline enables systematic analysis of thousands of units and their semantic selectivity.

Abstract

Convolutional neural networks (CNNs) learn abstract features to perform object classification, but understanding these features remains challenging due to difficult-to-interpret results or high computational costs. We propose an automatic method to visualize and systematically analyze learned features in CNNs. Specifically, we introduce a linking network that maps the penultimate layer of a pre-trained classifier to the latent space of a generative model (StyleGAN-XL), thereby enabling an interpretable, human-friendly visualization of the classifier's representations. Our findings indicate a congruent semantic order in both spaces, enabling a direct linear mapping between them. Training the linking network is computationally inexpensive and decoupled from training both the GAN and the classifier. We introduce an automatic pipeline that utilizes such GAN-based visualizations to quantify learned representations by analyzing activation changes in the classifier in the image domain. This quantification allows us to systematically study the learned representations in several thousand units simultaneously and to extract and visualize units selective for specific semantic concepts. Further, we illustrate how our method can be used to quantify and interpret the classifier's decision boundary using counterfactual examples. Overall, our method offers systematic and objective perspectives on learned abstract representations in CNNs. https://github.com/kaschube-lab/LinkingInStyle.git