A matter of attitude: Focusing on positive and active gradients to boost saliency maps

Oscar Llorente, Jaime Boal, Eugenio F. Sánchez-Úbeda
9/22/2023
cs.AIcs.LG

No official code found. Best implementation: osllogon/positive_active_saliency_maps (1 star, MIT, last commit 2025-01-02).

Abstract

Saliency maps have become one of the most widely used interpretability techniques for convolutional neural networks (CNN) due to their simplicity and the quality of the insights they provide. However, there are still some doubts about whether these insights are a trustworthy representation of what CNNs use to come up with their predictions. This paper explores how rescuing the sign of the gradients from the saliency map can lead to a deeper understanding of multi-class classification problems. Using both pretrained and trained from scratch CNNs we unveil that considering the sign and the effect not only of the correct class, but also the influence of the other classes, allows to better identify the pixels of the image that the network is really focusing on. Furthermore, how occluding or altering those pixels is expected to affect the outcome also becomes clearer.

View on arXivView PDF

Code Implementations(1)

osllogon/positive_active_saliency_mapsCommunity71%
MIT

Official implementation of "A matter of attitude: Focusing on positive and active gradients to boost saliency maps"

10Shell, PythonJul 26, 20221 years agoMIT
aiconvolutional-neural-networksdeep-learningexplainabilityexplainable-ai+2 more

Similar Papers

Training Better Deep Learning Models Using Human Saliency

85% match2024

Saliency Methods are Encoders: Analysing Logical Relations Towards Interpretation

84% match2024

A Learning Paradigm for Interpretable Gradients

84% match2024

SAGE: Saliency-Guided Contrastive Embeddings

84% match2025

Training for Trustworthy Saliency Maps: Adversarial Training Meets Feature-Map Smoothing

83% match2026

Increasing Interpretability of Neural Networks By Approximating Human Visual Saliency

83% match2024

Cite this paper

@article{llorente2023matter,
  title  = {A matter of attitude: Focusing on positive and active gradients to boost saliency maps},
  author = {Oscar Llorente and Jaime Boal and Eugenio F. Sánchez-Úbeda},
  year   = {2023},
  eprint = {2309.12913},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2309.12913v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2309.12913.svg)](https://codeofpaper.com/paper/2309.12913)

Discussion