Investigating the influence of noise and distractors on the interpretation of neural networks

TL;DR

This paper explores how noise and distractors affect neural network explanations, providing new theoretical insights to improve the selection of explanation models within the deep-Taylor decomposition framework.

Contribution

It introduces an analysis of the impact of noise and distractors on neural network explanations, filling a gap in understanding their influence on interpretability methods.

Findings

Noise and distractors significantly alter explanation results

Provides theoretical insights for selecting appropriate explanation models

Enhances understanding of explanation robustness in noisy environments

Abstract

Understanding neural networks is becoming increasingly important. Over the last few years different types of visualisation and explanation methods have been proposed. However, none of them explicitly considered the behaviour in the presence of noise and distracting elements. In this work, we will show how noise and distracting dimensions can influence the result of an explanation model. This gives a new theoretical insights to aid selection of the most appropriate explanation model within the deep-Taylor decomposition framework.