Revisiting Pathologies of Neural Models under Input Reduction

Canasai Kruengkrai, Junichi Yamagishi


Abstract
We revisit the question of why neural models tend to produce high-confidence predictions on inputs that appear nonsensical to humans. Previous work has suggested that the models fail to assign low probabilities to such inputs due to model overconfidence. We evaluate various regularization methods on fact verification benchmarks and find that this problem persists even with well-calibrated or underconfident models, suggesting that overconfidence is not the only underlying cause. We also find that regularizing the models with reduced examples helps improve interpretability but comes with the cost of miscalibration. We show that although these reduced examples are incomprehensible to humans, they can contain valid statistical patterns in the dataset utilized by the model.
Anthology ID:
2023.findings-acl.730
Volume:
Findings of the Association for Computational Linguistics: ACL 2023
Month:
July
Year:
2023
Address:
Toronto, Canada
Editors:
Anna Rogers, Jordan Boyd-Graber, Naoaki Okazaki
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
11504–11517
Language:
URL:
https://aclanthology.org/2023.findings-acl.730
DOI:
10.18653/v1/2023.findings-acl.730
Bibkey:
Cite (ACL):
Canasai Kruengkrai and Junichi Yamagishi. 2023. Revisiting Pathologies of Neural Models under Input Reduction. In Findings of the Association for Computational Linguistics: ACL 2023, pages 11504–11517, Toronto, Canada. Association for Computational Linguistics.
Cite (Informal):
Revisiting Pathologies of Neural Models under Input Reduction (Kruengkrai & Yamagishi, Findings 2023)
Copy Citation:
PDF:
https://aclanthology.org/2023.findings-acl.730.pdf