[Reading a thesis every week] Continual learning for named entity recognition(2021)

Amazon과 University of Illinois at Chicago(UIC)에서 얼마전 발표한 논문이다.

 

https://www.amazon.science/publications/continual-learning-for-named-entity-recognition

Continual learning for named entity recognition

 

Abstract

Named Entity Recognition (NER) is a vital task in various NLP applications. However, in many real-world scenarios (e.g., voice-enabled assistants) new named entity types are frequently introduced, entailing re-training NER models to support these new entity types. Re-annotating the original training data for the new entity types could be costly or even impossible when storage limitations or security concerns restrict access to that data, and annotating a new dataset for all of the entities becomes impractical and error-prone as the number of types increases. To tackle this problem, we introduce a novel Continual Learning approach for NER, which requires new training material to be annotated only for the new entity types. To preserve the existing knowledge previously learned by the model, we exploit the Knowledge Distillation (KD) framework, where the existing NER model acts as the teacher for a new NER model (i.e., the student), which learns the new entity types by using the new training material and retains knowledge of old entities by imitating the teacher’s outputs on this new training set. Our experiments show that this approach allows the student model to “progressively” learn to identify new entity types without forgetting the previously learned ones. We also present a comparison with multiple strong baselines to demonstrate that our approach is superior for continually updating an NER model.

 

 

• What is Teacher Student framework?

Torrey and Taylor introduced a teacher-student framework for reinforcement learning. In their framework, a student agent learns to perform a task through reinforcement learning, while a teacher agent may suggest an action in order to help the student learn faster. A few assumptions are made in this framework.

 

 

 

➞ Continual Learning(CL)을 Named Entity Recognition 태스크에 적용시킨 방법이다. 라벨링된 최소의 데이터셋으로 모델을 업데이트시키며 새로운 entity types을 식별해낼 수 있다고 한다. 

➞ 프라이버시 문제나 용량 제한 문제 등으로 이전 학습 데이터가 더 이상 유용하지 않은 경우나, multiple entity types을 위한 data annotating과정에서 많은 비용이 요구되는 경우에 유용한 방법이라고 소개하고 있다.

➞ 한 줄로 요약하자면, '망각(forgetting) 문제를 방지하면서 새로운 entity types을 스탭 바이 스탭으로 학습해나가는 방식'이라고 할 수 있다.

 

 

 

 

ContinualLearning_For_NamedEntityRecognition.pdf

0.56MB