Unlocking Feature Visualization for Deeper Networks with Magnitude Constrained Optimization - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

Unlocking Feature Visualization for Deeper Networks with Magnitude Constrained Optimization

Résumé

Feature visualization has gained substantial popularity, particularly after the influential work by Olah et al. in 2017, which established it as a crucial tool for explainability. However, its widespread adoption has been limited due to a reliance on tricks to generate interpretable images, and corresponding challenges in scaling it to deeper neural networks. Here, we describe MACO, a simple approach to address these shortcomings. The main idea is to generate images by optimizing the phase spectrum while keeping the magnitude constant to ensure that generated explanations lie in the space of natural images. Our approach yields significantly better results-both qualitatively and quantitatively-and unlocks efficient and interpretable feature visualizations for large state-of-the-art neural networks. We also show that our approach exhibits an attribution mechanism allowing us to augment feature visualizations with spatial importance. We validate our method on a novel benchmark for comparing feature visualization methods, and release its visualizations for all classes of the ImageNet dataset on Lens. Overall, our approach unlocks, for the first time, feature visualizations for large, state-of-the-art deep neural networks without resorting to any parametric prior image model.
Fichier principal
Vignette du fichier
2306.06805.pdf (20.97 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04391346 , version 1 (17-01-2024)

Identifiants

Citer

Thomas Fel, Thibaut Boissin, Victor Boutin, Agustin Picard, Paul Novello, et al.. Unlocking Feature Visualization for Deeper Networks with Magnitude Constrained Optimization. Conference on Neural Information Processing Systems (NeurIPS), 2023, Dec 2023, New Orleans (Louisiana), United States. ⟨hal-04391346⟩
12 Consultations
1 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More