From Selection to Generation: A Survey of LLM-based Active Learning

Active Learning (AL) has been a powerful paradigm for improving model efficiency and performance by selecting the most informative data points for labeling and training. In recent active learning frameworks, Large Language Models (LLMs) have been employed not only for selection but also for generating entirely new data instances and providing more cost-effective annotations. Motivated by the increasing importance of high-quality data and efficient model training in the era of LLMs, we present a comprehensive survey on LLM-based Active Learning. We introduce an intuitive taxonomy that categorizes these techniques and discuss the transformative roles LLMs can play in the active learning loop. We further examine the impact of AL on LLM learning paradigms and its applications across various domains. Finally, we identify open challenges and propose future research directions. This survey aims to serve as an up-to-date resource for researchers and practitioners seeking to gain an intuitive understanding of LLM-based AL techniques and deploy them to new applications. * Equal contributions Model Training LLM-based Selection / Generation selected new generated 𝑥 ! data point data point LLM-based Annotation or Human Annotation initial data u n la b e le d d a ta 𝑥 " selected/gen. data points (𝑥 " , 𝑦 " ) (𝑥 ! , 𝑦 ! ) Class General Mechanism Description Querying (Section 3) Traditional Selection (Sec. 3.1) This class of techniques uses traditional selection such as uncertainty sampling, disagreement, gradient-based sampling, and so on. LLM-based Selection (Sec. 3.2) The class of LLM-based selection techniques focus on using LLMs to select the instances. LLM-based Generation (Sec. 3.3) The class of LLM-based generation techniques focus on generating novel instances. Hybrid (Sec. 3.4) Combines advantages of both LLM-based selection and generation Annotation (Section 4) Human Annotation (Sec. 4.1) Traditional human annotation simply refers to using humans to annotate the selected or generated instances, which is costly. LLM-based Annotation (Sec. 4.2) The class of LLM-based annotation techniques focus on leveraging LLMs for annotation and evaluation. This class of techniques are far cheaper than human annotation. Hybrid (Sec. 4.3) This class of techniques aim to leverage the advantages of both humans and LLMs for optimal annotations while minimizing cost