Difference-Masking: Choosing What to Mask in Continued Pretraining

Alex Wilf, Syeda Akter, Leena Mathur, Paul Liang, Sheryl Mathew, Mengrou Shou, Eric Nyberg, Louis-Philippe Morency


Abstract
The self-supervised objective of masked prediction has led to promising performance gains on a variety of downstream tasks. However, while most approaches randomly mask tokens, there is strong intuition that deciding what to mask can substantially improve learning outcomes. We investigate this in continued pretraining setting in which pretrained models continue to pretrain on domain-specific data before performing some downstream task. We introduce Difference-Masking, a masking strategy that automatically chooses what to mask during continued pretraining by considering what makes a task domain different from the pretraining domain. Empirically, we find that Difference-Masking outperforms baselines on continued pretraining settings across four diverse language-only and multimodal video tasks.
Anthology ID:
2023.findings-emnlp.881
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2023
Month:
December
Year:
2023
Address:
Singapore
Editors:
Houda Bouamor, Juan Pino, Kalika Bali
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
13222–13234
Language:
URL:
https://aclanthology.org/2023.findings-emnlp.881
DOI:
10.18653/v1/2023.findings-emnlp.881
Bibkey:
Cite (ACL):
Alex Wilf, Syeda Akter, Leena Mathur, Paul Liang, Sheryl Mathew, Mengrou Shou, Eric Nyberg, and Louis-Philippe Morency. 2023. Difference-Masking: Choosing What to Mask in Continued Pretraining. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 13222–13234, Singapore. Association for Computational Linguistics.
Cite (Informal):
Difference-Masking: Choosing What to Mask in Continued Pretraining (Wilf et al., Findings 2023)
Copy Citation:
PDF:
https://preview.aclanthology.org/ingest-2024-clasp/2023.findings-emnlp.881.pdf