Pool-Based Sequential Active Learning For Regression Based on Incremental Cluster Center Selection
Binwei Lin, Zhiyong Yu, Fangwan Huang, Longkun Guo · 2022
Active learning uses a certain algorithm to select the most useful samples for labeling to obtain a model with better performance, thereby reducing labeling costs. Active learning can be used not only for classification, but also for regression, named as ALR. Pool-based sequential active learning refers to selecting one sample from a pool for labeling in each iteration. For pool-based sequential ALR, existing methods are guided by criteria such as informativeness, representativeness, or diversity singly, but rarely consider these three criteria synthetically. In this paper, we propose a pool-based sequential ALR method based on incremental cluster center selection (ICS), which takes into account the representativeness and diversity. Besides, we further improve ICS by combining it with informativeness-based methods, so that the combined method comprehensively considers the above three criteria. Experiments on multiple real world datasets in various fields, using ridge regression, proved the effectiveness of the proposed methods.