Do Feature Attribution Methods Correctly Attribute Features?
This addresses the problem of unreliable interpretability in machine learning for researchers and practitioners, but it is incremental as it builds on existing evidence.
The paper tackles the lack of consensus and ground truth in evaluating feature attribution methods by proposing a dataset modification procedure to induce ground truth, and finds that common methods like saliency maps, rationales, and attentions have deficiencies, questioning their correctness and reliability.
Feature attribution methods are popular in interpretable machine learning. These methods compute the attribution of each input feature to represent its importance, but there is no consensus on the definition of "attribution", leading to many competing methods with little systematic evaluation, complicated in particular by the lack of ground truth attribution. To address this, we propose a dataset modification procedure to induce such ground truth. Using this procedure, we evaluate three common methods: saliency maps, rationales, and attentions. We identify several deficiencies and add new perspectives to the growing body of evidence questioning the correctness and reliability of these methods applied on datasets in the wild. We further discuss possible avenues for remedy and recommend new attribution methods to be tested against ground truth before deployment. The code is available at https://github.com/YilunZhou/feature-attribution-evaluation