arXiv Analytics

Sign in

arXiv:2103.13533 [cs.LG]AbstractReferencesReviewsResources

Symmetry-Preserving Paths in Integrated Gradients

Miguel Lerma, Mirtha Lucas

Published 2021-03-25Version 1

We provide rigorous proofs that the Integrated Gradients (IG) attribution method for deep networks satisfies completeness and symmetry-preserving properties. We also study the uniqueness of IG as a path method preserving symmetry.

Comments: 7 pages, 1 figure
Categories: cs.LG, cs.AI
Subjects: 68T07, I.2.m
Related articles: Most relevant | Search more
arXiv:2307.03305 [cs.LG] (Published 2023-07-06)
A Vulnerability of Attribution Methods Using Pre-Softmax Scores
arXiv:2310.04821 [cs.LG] (Published 2023-10-07, updated 2023-10-10)
Rethink Baseline of Integrated Gradients from the Perspective of Shapley Value
Shuyang Liu et al.
arXiv:2204.05948 [cs.LG] (Published 2022-04-12)
Maximum Entropy Baseline for Integrated Gradients

Links

Toolbox