LaFAM: Unsupervised Feature Attribution with Label-free Activation Maps

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Karjauv, Aray, Albayrak, Sahin
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910900647100416
author Karjauv, Aray
Albayrak, Sahin
author_facet Karjauv, Aray
Albayrak, Sahin
contents Convolutional Neural Networks (CNNs) are known for their ability to learn hierarchical structures, naturally developing detectors for objects, and semantic concepts within their deeper layers. Activation maps (AMs) reveal these saliency regions, which are crucial for many Explainable AI (XAI) methods. However, the direct exploitation of raw AMs in CNNs for feature attribution remains underexplored in literature. This work revises Class Activation Map (CAM) methods by introducing the Label-free Activation Map (LaFAM), a streamlined approach utilizing raw AMs for feature attribution without reliance on labels. LaFAM presents an efficient alternative to conventional CAM methods, demonstrating particular effectiveness in saliency map generation for self-supervised learning while maintaining applicability in supervised learning scenarios.
format Preprint
id arxiv_https___arxiv_org_abs_2407_06059
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LaFAM: Unsupervised Feature Attribution with Label-free Activation Maps
Karjauv, Aray
Albayrak, Sahin
Computer Vision and Pattern Recognition
Convolutional Neural Networks (CNNs) are known for their ability to learn hierarchical structures, naturally developing detectors for objects, and semantic concepts within their deeper layers. Activation maps (AMs) reveal these saliency regions, which are crucial for many Explainable AI (XAI) methods. However, the direct exploitation of raw AMs in CNNs for feature attribution remains underexplored in literature. This work revises Class Activation Map (CAM) methods by introducing the Label-free Activation Map (LaFAM), a streamlined approach utilizing raw AMs for feature attribution without reliance on labels. LaFAM presents an efficient alternative to conventional CAM methods, demonstrating particular effectiveness in saliency map generation for self-supervised learning while maintaining applicability in supervised learning scenarios.
title LaFAM: Unsupervised Feature Attribution with Label-free Activation Maps
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2407.06059