GAM: Explainable Visual Similarity and Classification via Gradient Activation Maps

TL;DR

GAM introduces a novel method for explaining visual similarity and classification model predictions by utilizing localized gradient and activation information from multiple network layers, providing more accurate visual explanations.

Contribution

GAM is a new explainability technique that leverages multi-layer gradient and activation data to improve visual explanations over existing methods.

Findings

GAM outperforms existing explanation methods across various datasets.

GAM provides more localized and accurate visual explanations.

Empirical validation confirms GAM's effectiveness in different tasks.

Abstract

We present Gradient Activation Maps (GAM) - a machinery for explaining predictions made by visual similarity and classification models. By gleaning localized gradient and activation information from multiple network layers, GAM offers improved visual explanations, when compared to existing alternatives. The algorithmic advantages of GAM are explained in detail, and validated empirically, where it is shown that GAM outperforms its alternatives across various tasks and datasets.