Towards a Unified Evaluation of Explanation Methods without Ground Truth

Hao Zhang, Jiayi Chen, Haotian Xue, Quanshi Zhang

Keywords: Deep Learning, Interpretability, Evaluation, Convolutional Neural Networks

Abstract:

This paper proposes a set of criteria to evaluate the objectiveness of explanation
methods of neural networks, which is crucial for the development of explainable
AI, but it also presents significant challenges. The core challenge is that people
usually cannot obtain ground-truth explanations of the neural network. To this
end, we design four metrics to evaluate the explanation result without ground-truth
explanations. Our metrics can be broadly applied to nine benchmark methods of
interpreting neural networks, which provides new insights of explanation methods.