The Impact of Integration Step on Integrated Gradients

Masahiro Makino, Yuya Asazuma, Shota Sasaki, Jun Suzuki


Abstract
Integrated Gradients (IG) serve as a potent tool for explaining the internal structure of a language model. The calculation of IG requires numerical integration, wherein the number of steps serves as a critical hyperparameter. The step count can drastically alter the results, inducing considerable errors in interpretability. To scrutinize the effect of step variation on IG, we measured the difference between theoretical and observed IG totals for each step amount.Our findings indicate that the ideal number of steps to maintain minimal error varies from instance to instance. Consequently, we advocate for customizing the step count for each instance. Our study is the first to quantitatively analyze the variation of IG values with the number of steps.
Anthology ID:
2024.eacl-srw.22
Volume:
Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: Student Research Workshop
Month:
March
Year:
2024
Address:
St. Julian’s, Malta
Editors:
Neele Falk, Sara Papi, Mike Zhang
Venue:
EACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
279–289
Language:
URL:
https://aclanthology.org/2024.eacl-srw.22
DOI:
Bibkey:
Cite (ACL):
Masahiro Makino, Yuya Asazuma, Shota Sasaki, and Jun Suzuki. 2024. The Impact of Integration Step on Integrated Gradients. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: Student Research Workshop, pages 279–289, St. Julian’s, Malta. Association for Computational Linguistics.
Cite (Informal):
The Impact of Integration Step on Integrated Gradients (Makino et al., EACL 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.eacl-srw.22.pdf