Synergistic Training: Harnessing Active Learning and Pseudo-Labeling for Enhanced Model Performance in Deep Learning

Abstract

This research addresses the growing need for efficient data labeling methods by leveraging deep learning models. The proposed approach combines pre-training and active learning to automate the labeling process and reduce reliance on human annotators. In the pre-training phase, two deep learning models are trained using labeled data, adjusting the data ratio to ensure approximately 50% accuracy on the test set. In the active learning phase, the models generate pseudo labels for unlabeled data based on a confidence threshold, and the selected data is used to improve the models' performance through alternating epochs. The experimental results demonstrate the effectiveness of the approach, achieving significant improvements in accuracy compared to traditional methods. This research contributes to the trend of using deep learning for efficient data labeling and offers a promising solution for reducing the time and cost associated with manual annotation.