Query By Bagging

Query By Bagging is an active learning technique used in machine learning to select the most informative data points for labeling from a large unlabeled dataset. It works by training multiple models (a 'bagging' ensemble) on the labeled data and using their disagreement or uncertainty to identify instances where the models are most uncertain, which are then prioritized for human annotation. This approach aims to reduce labeling costs and improve model performance by focusing on the most valuable data.

Also known as: QBB, Query-by-Bagging, Active Learning with Bagging, Bagging-based Query Strategy, Uncertainty Sampling with Bagging

Why learn Query By Bagging?

Developers should learn Query By Bagging when building machine learning models with limited labeled data, such as in natural language processing, computer vision, or medical diagnosis applications, to efficiently allocate labeling resources. It is particularly useful in scenarios where labeling is expensive or time-consuming, as it helps train more accurate models with fewer labeled examples by selecting data that reduces model uncertainty. This methodology is often applied in semi-supervised learning settings to enhance data efficiency and model robustness.

Learning Resources

• book: Active Learning Literature Survey
• docs: Scikit-learn Active Learning Documentation
• tutorial: Active Learning Tutorial on Towards Data Science
• docs: Query by Bagging Paper by Abe and Mamitsuka