Research work · 2025
Active Learning Strategies for Optimising Data Annotation Using Machine Learning
V. Hariharan, K. Sivaraman
Machine Learning, Predictive Analytics, and Optimization in Complex Systems
Abstract
This chapter investigates the potential of active learning techniques, in particular Bayesian Optimization, to enhance data annotation and model training. Active learning selects meaningful examples from unlabeled data to improve model performance and reduce labeling effort. Several active learning techniques, including uncertainty sampling, query-by-committee, and Bayesian Optimization, are investigated to determine how this impacts data annotation and model training. Multiple datasets, including CIFAR-10, IMDB reviews, and COCO, have been evaluated experimentally to demonstrate that Bayesian Optimization improves accuracy, reduces annotation effort, and accelerates model convergence. The findings emphasize the potential of active learning motivated by Bayesian optimization for efficient machine learning with limited resources. According to the findings, active learning strategies, hybrid approaches, and their use in real-world, ever-changing contexts warrant additional research.