Rethinking Perturbations in Encoder-Decoders for Fast Training

Sho Takase, Shun Kiyono


Abstract
We often use perturbations to regularize neural models. For neural encoder-decoders, previous studies applied the scheduled sampling (Bengio et al., 2015) and adversarial perturbations (Sato et al., 2019) as perturbations but these methods require considerable computational time. Thus, this study addresses the question of whether these approaches are efficient enough for training time. We compare several perturbations in sequence-to-sequence problems with respect to computational time. Experimental results show that the simple techniques such as word dropout (Gal and Ghahramani, 2016) and random replacement of input tokens achieve comparable (or better) scores to the recently proposed perturbations, even though these simple methods are faster.
Anthology ID:
2021.naacl-main.460
Volume:
Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies
Month:
June
Year:
2021
Address:
Online
Venue:
NAACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
5767–5780
Language:
URL:
https://aclanthology.org/2021.naacl-main.460
DOI:
10.18653/v1/2021.naacl-main.460
Bibkey:
Cite (ACL):
Sho Takase and Shun Kiyono. 2021. Rethinking Perturbations in Encoder-Decoders for Fast Training. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5767–5780, Online. Association for Computational Linguistics.
Cite (Informal):
Rethinking Perturbations in Encoder-Decoders for Fast Training (Takase & Kiyono, NAACL 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.naacl-main.460.pdf
Video:
 https://aclanthology.org/2021.naacl-main.460.mp4
Code
 takase/rethink_perturbations
Data
DUC 2004WMT 2014