Logic Traps in Evaluating Attribution Scores
arXiv:2109.05463
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
arXiv comment12 pages, accepted by ACL 2022
References in corpus (17)
- Towards A Rigorous Science of Interpretable Machine Learning
- Axiomatic Attribution for Deep Networks
- Learning Important Features Through Propagating Activation Differences
- Attention is not Explanation
- Understanding Neural Networks through Representation Erasure
- Beyond Word Importance: Contextual Decomposition to Extract Interactions from LSTMs
- On the Robustness of Interpretability Methods
- Attention Interpretability Across NLP Tasks
- L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data
- How does this interaction affect me? Interpretable attribution for feature interactions
- Self-Attention Attribution: Interpreting Information Interactions Inside Transformer
- AllenNLP Interpret: A Framework for Explaining Predictions of NLP Models
- 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
- Gradient-based Analysis of NLP Models is Manipulable
- The elephant in the interpretability room: Why use attention as explanation when we have saliency methods?