A batch ensemble approach to active learning with model selection.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computer Science

Background:

• Supervised learning relies on optimal training data selection (active learning) and model choice (model selection).
• These crucial components have historically been addressed independently, potentially limiting overall performance.
• Simultaneous optimization of both aspects is hypothesized to yield superior generalization capabilities.

Purpose of the Study:

• To introduce a novel approach, ensemble active learning, that integrates active learning and model selection.
• To demonstrate the effectiveness of simultaneously optimizing training input points and model selection.
• To improve the generalization performance of supervised learning models.

Main Methods:

• Developed a new method named ensemble active learning.
• Integrated the processes of active learning and model selection into a single framework.
• Conducted numerical experiments to validate the proposed approach.

Main Results:

• The proposed ensemble active learning method achieved favorable results.
• Demonstrated superior performance compared to alternative sequential approaches.
• Showcased the benefits of joint optimization of data selection and model choice.

Conclusions:

• Ensemble active learning effectively addresses both active learning and model selection concurrently.
• Simultaneous optimization offers significant advantages over sequential methods in supervised learning.
• The proposed approach provides a more efficient and effective strategy for improving model generalization.