Logic Traps in Evaluating Attribution Scores

Abstract

Modern deep learning models are notoriously opaque, which has motivated the development of methods for interpreting how deep models predict. This goal is usually approached with attribution method, which assesses the influence of features on model predictions. As an explanation method, the evaluation criteria of attribution methods is how accurately it re-reflects the actual reasoning process of the model (faithfulness). Meanwhile, since the reasoning process of deep models is inaccessible, researchers design various evaluation methods to demonstrate their arguments. However, some crucial logic traps in these evaluation methods are ignored in most works, causing inaccurate evaluation and unfair comparison. This paper systematically reviews existing methods for evaluating attribution scores and summarizes the logic traps in these methods. We further conduct experiments to demonstrate the existence of each logic trap. Through both the theoretical and experimental analysis, we hope to increase attention on the inaccurate evaluation of attribution scores. Moreover, with this paper, we suggest stopping focusing on improving performance under unreliable evaluation systems and starting efforts on reducing the impact of proposed logic traps

12 pages, accepted by ACL 2022

References in corpus (8)

• Towards A Rigorous Science of Interpretable Machine Learning
• Attention is not Explanation
• Understanding Neural Networks through Representation Erasure
• Attention Interpretability Across NLP Tasks
• How does this interaction affect me? Interpretable attribution for feature interactions
• Interpretation of NLP models through input marginalization
• Evaluating Saliency Methods for Neural Language Models
• Learning Variational Word Masks to Improve the Interpretability of Neural Text Classifiers