Investigating the influence of noise and distractors on the interpretation of neural networks
This work addresses the reliability of neural network explanations for researchers and practitioners, but it is incremental as it builds on existing deep-Taylor decomposition methods.
The paper tackled the problem of how noise and distracting elements affect the interpretation of neural networks, showing that these factors influence explanation models and providing theoretical insights for selecting appropriate models within the deep-Taylor decomposition framework.
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.