Continual Active Learning with Ensemble Method

Machine learning (ML) and deep learning (DL) have shown significant advancements in recent years, but their application in real field often encounters challenges due to insufficient labeled data. This labeling process, especially in specialized domains requiring expert knowledge, demands expert input, making it costly and time-consuming. Active learning (AL) method addresses this challenge by iteratively selecting the most informative data points for labeling, thus reducing the labeling burden. This paper proposes an enhanced AL method that employs ensemble techniques to improve the selection of critical data points. By combining multiple data extraction methods, our approach identifies the most valuable samples for labeling more effectively. The effectiveness of the proposed method is validated through experiments on a binary classification dataset from computer numerical control (CNC) milling machine and the multi-class modified national institute of standards and technology (MNIST) dataset. The results demonstrate that our Active Learning with Ensemble Method (ALEM) significantly enhances labeling efficiency and model performance, providing a robust solution for scenarios with limited labeled data.