Causal Feature Attribution: Towards a Trustworthy and Actionable Explanations of Deep Neural Network

Zhixuan Chu, Mengxuan Hu, Siqiao Xue, Qing Cui, Longfei Li, Sheng Li

Keywords: explainable artificial intelligence

Abstract: Nowadays, deep learning-based models have shown extraordinary performance on various tasks; however, the most significant bottleneck is the lack of transparency and explainability. Although many Explainable Artificial Intelligence (XAI) models have been proposed to provide feature attributions and generate explanations for back-box models, most of them are correlation-based methods, which fail to provide trustworthy explanations due to the misleading correlations learned by black-box models. Therefore, we propose a causality-based feature attribution model that can provide trustworthy explanations by treating the feature of interest as an intervention and inferring its causal feature attribution. Extensive results show that our model can provide trustworthy and actionable explanations.

Primary Area: general machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8814