A General Feature Attribution Framework under a Black-box Setting

ICLR 2025 Conference Submission13816 Authors

Keywords: Explainability, Feature Attribution, Black-box Setting, Query-level Access

Abstract:

Feature attribution is widely accepted as a form of explanation for reasoning machine decisions, indicating the proportion of each feature's contribution to an inquired decision.
While most efforts have focused on determining attributions through exact gradient measurements, recent work has adopted gradient estimation to derive explanatory information requiring only query-level access – a restricted yet more practical accessibility assumption known as the black-box setting.
Following this direction, this paper extends the idea of utilizing estimated gradients to a broader framework and introduces GEFA (Gradient-Estimation-based explanation For All). Unlike the previous attempt that focused on explaining image classifiers, the proposed explainer derives feature attributions in a proxy space, making it generally applicable to arbitrary black-box models, regardless of input type.
In addition to its close relationship with Integrated Gradients, we find, surprisingly, that our approach – a path method built upon estimated gradients – outputs unbiased estimates of Shapley Values. By avoiding the potential information waste sourced from computing marginal contributions, it improves the quality of derived explanations, as demonstrated by our quantitative evaluations.

Primary Area: interpretability and explainable AI

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/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 13816