On the Importance of Effectively Adapting Pretrained Language Models for Active Learning

2021/04/16 by Katerina Margatina, Margatina, Katerina, Loïc Barrault +4 · 4 citations
Computer Science · Engineering · #Active learning (machine learning) #Artificial intelligence #Computation and Language (cs.CL) #Computer science #Engineering #FOS: Computer and information sciences #Labeled data #Language model #Machine Learning and Algorithms #Machine learning #Natural Language Processing Techniques #Simple (philosophy) #Task (project management) #Topic Modeling #Training set #cs.CL

paper · pdf · doi:10.48550/arxiv.2104.08320

published in arXiv (Cornell University) (Cornell University) · To appear at ACL 2022

openalex publication_date 2021/04/16 · arxiv created 2022/03/02 · arxiv updated 2022/03/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

Recent Active Learning (AL) approaches in Natural Language Processing (NLP) proposed using off-the-shelf pretrained language models (LMs). In this paper, we argue that these LMs are not adapted effectively to the downstream task during AL and we explore ways to address this issue. We suggest to first adapt the pretrained LM to the target task by continuing training with all the available unlabeled data and then use it for AL. We also propose a simple yet effective fine-tuning method to ensure that the adapted LM is properly trained in both low and high resource scenarios during AL. Our experiments demonstrate that our approach provides substantial data efficiency improvements compared to the standard fine-tuning approach, suggesting that a poor training strategy can be catastrophic for AL.

Cited by

• Op-Fed: Opinion, Stance, and Monetary Policy Annotations on FOMC Transcripts Using Active Learning
• Generalized Category Discovery in Event-Centric Contexts: Latent Pattern Mining with LLMs
• No Free Lunch in Active Learning: LLM Embedding Quality Dictates Query Strategy Success
• Active Learning Methods for Efficient Data Utilization and Model Performance Enhancement

Related

• Towards Efficient Active Learning in NLP via Pretrained Representations
• From Selection to Generation: A Survey of LLM-based Active Learning
• Improving Language Plasticity via Pretraining with Active Forgetting
• A Survey of Active Learning for Natural Language Processing
• Smooth Sailing: Improving Active Learning for Pre-trained Language Models with Representation Smoothness Analysis