Abstract
In this paper, we study trade-offs between efficiency, cost and accuracy when pre-training Transformer encoders with different pre-training objectives. For this purpose, we analyze features of common objectives and combine them to create new effective pre-training approaches. Specifically, we designed light token generators based on a straightforward statistical approach, which can replace ELECTRA computationally heavy generators, thus highly reducing cost. Our experiments also show that (i) there are more efficient alternatives to BERT’s MLM, and (ii) it is possible to efficiently pre-train Transformer-based models using lighter generators without a significant drop in performance.- Anthology ID:
- 2022.findings-emnlp.405
- Volume:
- Findings of the Association for Computational Linguistics: EMNLP 2022
- Month:
- December
- Year:
- 2022
- Address:
- Abu Dhabi, United Arab Emirates
- Editors:
- Yoav Goldberg, Zornitsa Kozareva, Yue Zhang
- Venue:
- Findings
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 5533–5547
- Language:
- URL:
- https://aclanthology.org/2022.findings-emnlp.405
- DOI:
- 10.18653/v1/2022.findings-emnlp.405
- Cite (ACL):
- Luca Di Liello, Matteo Gabburo, and Alessandro Moschitti. 2022. Effective Pretraining Objectives for Transformer-based Autoencoders. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 5533–5547, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
- Cite (Informal):
- Effective Pretraining Objectives for Transformer-based Autoencoders (Di Liello et al., Findings 2022)
- PDF:
- https://preview.aclanthology.org/alta-23-ingestion/2022.findings-emnlp.405.pdf