Diversity-Based Active Learning: An Evaluation of Metric Spaces for Active Learning Selection
2026-08-24 • Machine Learning
Machine Learning
AI summaryⓘ
The authors study a way to save effort when teaching computers to recognize things by smartly choosing which examples to label. They focus on a method called Greedy K-center that picks diverse examples from different types of data views, like raw features or probability predictions. Their tests show that using the computer’s prediction probabilities, especially when adjusted by how uncertain the model is (entropy), works better than other ways of picking examples. This helps reduce the number of labeled examples needed to train good classifiers.
Active LearningGreedy K-centerDiversity-based selectionLinear Discriminant Analysis (LDA)Random ForestEntropy weightingUnlabeled dataClassificationProbability space
Authors
Siddharth Chilamkur, Dorit S. Hochbaum
Abstract
With rapid advancement over the last few years, many different methods are now widely used for classification. However, training these models requires substantial labeled data. Active Learning is a potential solution to this problem. Pool-based active learning minimizes costs by querying only the most informative samples from an unlabeled dataset. Diversity-based approaches, on the other hand, attempt to select a representative subset of the data. There are many different objectives for determining the selection process, including exact K-center, exact K-median, and Greedy K-center. In this paper, we will focus on evaluating the performance of Greedy K-center across a variety of metric spaces: the raw feature space, a Linear Discriminant Analysis (LDA) space, and a model-derived probability space (with and without entropy-based weighting). Using Random Forest classifiers as a baseline evaluator, our empirical results on synthetic and real-world datasets demonstrate that mapping unlabeled instances into a predictive probability space and weighting the result by entropy often dominates the other options for active learning selection with Greedy K-center.