Controlling Explanatory Heatmap Resolution and Semantics via Decomposition Depth

TL;DR

This paper explores how adjusting the decomposition depth in Layer-wise Relevance Propagation affects the resolution and semantic content of heatmaps, aiding interpretability of deep classifiers.

Contribution

It demonstrates how varying the cutoff points in LRP influences heatmap resolution and semantics, comparing CNNs and Fisher Vector classifiers.

Findings

Heatmap resolution can be controlled via decomposition depth.

Semantic content of heatmaps varies with cutoff points.

LRP reveals different prediction strategies of classifiers.

Abstract

We present an application of the Layer-wise Relevance Propagation (LRP) algorithm to state of the art deep convolutional neural networks and Fisher Vector classifiers to compare the image perception and prediction strategies of both classifiers with the use of visualized heatmaps. Layer-wise Relevance Propagation (LRP) is a method to compute scores for individual components of an input image, denoting their contribution to the prediction of the classifier for one particular test point. We demonstrate the impact of different choices of decomposition cut-off points during the LRP-process, controlling the resolution and semantics of the heatmap on test images from the PASCAL VOC 2007 test data set.