Saved in:
Bibliographic Details
Main Authors: Rodrigues, Caroline Mazini, Boutry, Nicolas, Najman, Laurent
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2401.14434
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913350714130432
author Rodrigues, Caroline Mazini
Boutry, Nicolas
Najman, Laurent
author_facet Rodrigues, Caroline Mazini
Boutry, Nicolas
Najman, Laurent
contents The explication of Convolutional Neural Networks (CNN) through xAI techniques often poses challenges in interpretation. The inherent complexity of input features, notably pixels extracted from images, engenders complex correlations. Gradient-based methodologies, exemplified by Integrated Gradients (IG), effectively demonstrate the significance of these features. Nevertheless, the conversion of these explanations into images frequently yields considerable noise. Presently, we introduce GAD (Gradient Artificial Distancing) as a supportive framework for gradient-based techniques. Its primary objective is to accentuate influential regions by establishing distinctions between classes. The essence of GAD is to limit the scope of analysis during visualization and, consequently reduce image noise. Empirical investigations involving occluded images have demonstrated that the identified regions through this methodology indeed play a pivotal role in facilitating class differentiation.
format Preprint
id arxiv_https___arxiv_org_abs_2401_14434
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Transforming gradient-based techniques into interpretable methods
Rodrigues, Caroline Mazini
Boutry, Nicolas
Najman, Laurent
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
The explication of Convolutional Neural Networks (CNN) through xAI techniques often poses challenges in interpretation. The inherent complexity of input features, notably pixels extracted from images, engenders complex correlations. Gradient-based methodologies, exemplified by Integrated Gradients (IG), effectively demonstrate the significance of these features. Nevertheless, the conversion of these explanations into images frequently yields considerable noise. Presently, we introduce GAD (Gradient Artificial Distancing) as a supportive framework for gradient-based techniques. Its primary objective is to accentuate influential regions by establishing distinctions between classes. The essence of GAD is to limit the scope of analysis during visualization and, consequently reduce image noise. Empirical investigations involving occluded images have demonstrated that the identified regions through this methodology indeed play a pivotal role in facilitating class differentiation.
title Transforming gradient-based techniques into interpretable methods
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2401.14434