LaFAM: Unsupervised Feature Attribution with Label-free Activation Maps

TL;DR

LaFAM introduces a label-free, efficient method for feature attribution using raw activation maps in CNNs, enhancing saliency map generation especially in self-supervised learning without relying on labels.

Contribution

The paper proposes LaFAM, a novel approach that leverages raw activation maps for feature attribution without labels, improving efficiency and applicability over traditional CAM methods.

Findings

Effective in saliency map generation for self-supervised learning

Operates without reliance on labels, broadening applicability

Demonstrates efficiency over conventional CAM methods

Abstract

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.