Continual Machine Reading Comprehension via Uncertainty-aware Fixed Memory and Adversarial Domain Adaptation

Zhijing Wu, Hua Xu*, Jingliang Fang, Kai Gao

*此作品的通讯作者

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

Continual Machine Reading Comprehension aims to incrementally learn from a continuous data stream across time without access the previous seen data, which is crucial for the development of real-world MRC systems. However, it is a great challenge to learn a new domain incrementally without catastrophically forgetting previous knowledge. In this paper, MA-MRC, a continual MRC model with uncertainty-aware fixed Memory and Adversarial domain adaptation, is proposed. In MA-MRC, a fixed size memory stores a small number of samples in previous domain data along with an uncertainty-aware updating strategy when new domain data arrives. For incremental learning, MA-MRC not only keeps a stable understanding by learning both memory and new domain data, but also makes full use of the domain adaptation relationship between them by adversarial learning strategy. The experimental results show that MA-MRC is superior to strong baselines and has a substantial incremental learning ability without catastrophically forgetting under two different continual MRC settings.

源语言英语
主期刊名Findings of the Association for Computational Linguistics
主期刊副标题NAACL 2022 - Findings
出版商Association for Computational Linguistics (ACL)
2330-2339
页数10
ISBN(电子版)9781955917766
出版状态已出版 - 2022
已对外发布
活动2022 Findings of the Association for Computational Linguistics: NAACL 2022 - Seattle, 美国
期限: 10 7月 202215 7月 2022

出版系列

姓名Findings of the Association for Computational Linguistics: NAACL 2022 - Findings

会议

会议2022 Findings of the Association for Computational Linguistics: NAACL 2022
国家/地区美国
Seattle
时期10/07/2215/07/22

指纹

探究 'Continual Machine Reading Comprehension via Uncertainty-aware Fixed Memory and Adversarial Domain Adaptation' 的科研主题。它们共同构成独一无二的指纹。

引用此

Wu, Z., Xu, H., Fang, J., & Gao, K. (2022). Continual Machine Reading Comprehension via Uncertainty-aware Fixed Memory and Adversarial Domain Adaptation. 在 Findings of the Association for Computational Linguistics: NAACL 2022 - Findings (页码 2330-2339). (Findings of the Association for Computational Linguistics: NAACL 2022 - Findings). Association for Computational Linguistics (ACL).