MBA: Mini-Batch AUC Optimization.

Area of Science:

• Machine Learning
• Data Science
• Statistical Modeling

Background:

• Area Under the Curve (AUC) is a key metric in machine learning.
• Optimizing AUC is crucial but challenging for large datasets.
• Existing scalable AUC optimization methods face limitations with very large data.

Purpose of the Study:

• To propose a novel and efficient approach for AUC maximization.
• To address the challenge of handling very large datasets in AUC optimization.
• To develop a scalable method that is simple, fast, and learning-rate free.

Main Methods:

• Proposes AUC maximization via sampling mini-batches of positive/negative instance pairs.
• Utilizes U-statistics to approximate a global risk minimization problem.
• The algorithm is learning-rate free and computationally efficient.

Main Results:

• The number of samples needed for good performance is independent of the total number of pairs.
• Achieves efficient AUC maximization even for datasets with a quadratic number of instance pairs.
• Demonstrates practical utility and effectiveness through extensive experiments.

Conclusions:

• The proposed method offers a scalable and efficient solution for AUC maximization.
• It effectively handles large datasets where traditional methods struggle.
• The approach is practical and shows significant utility in real-world applications.