Open Dien-ES opened 2 years ago
@yukyunglee 님께서 추가로 소개해주신 같이 읽으면 좋은 논문
Recall and Learn: Fine-tuning Deep Pretrained Language Models with Less Forgetting
AVocaDo: Strategy for Adapting Vocabulary to Downstream Domain
Task-adaptive Pre-training and Self-training are Complementary for Natural Language Understanding
@jwkanggist 추가 추천한 논문
Just Rank: Rethinking Evaluation with Word and Sentence Similarities
Keywords
RoBERTa, Language model, Domain-adaptive pretraining, Task-adaptive pretraining
TL;DR
Multiphase adaptive pretraining with domain and task corpus offers large gains in task performance.
Abstract
Language models pretrained on text from a wide variety of sources form the foundation of today’s NLP. In light of the success of these broad-coverage models, we investigate whether it is still helpful to tailor a pretrained model to the domain of a target task. We present a study across four domains (biomedical and computer science publications, news, and reviews) and eight classification tasks, showing that a second phase of pretraining indomain (domain-adaptive pretraining) leads to performance gains, under both high- and low-resource settings. Moreover, adapting to the task’s unlabeled data (task-adaptive pretraining) improves performance even after domain-adaptive pretraining. Finally, we show that adapting to a task corpus augmented using simple data selection strategies is an effective alternative, especially when resources for domain-adaptive pretraining might be unavailable. Overall, we consistently find that multiphase adaptive pretraining offers large gains in task performance.
Paper link
https://aclanthology.org/2020.acl-main.740/
Presentation link
https://drive.google.com/file/d/1mziQEteSwHxLZ6Jb0vCsX3w1KpUGGmzA/view?usp=sharing
video link
https://youtu.be/2W_2vHamLYo