MBA: Mini-Batch AUC Optimization
This addresses scalability issues in AUC optimization for machine learning and signal processing applications, though it is incremental as it builds on existing scalable methods.
The paper tackles the challenge of optimizing AUC for very large datasets by proposing a mini-batch approach that samples positive/negative pairs and uses U-statistics, resulting in a simple, fast, and learning-rate-free algorithm with performance independent of the quadratic number of pairs.
Area under the receiver operating characteristics curve (AUC) is an important metric for a wide range of signal processing and machine learning problems, and scalable methods for optimizing AUC have recently been proposed. However, handling very large datasets remains an open challenge for this problem. This paper proposes a novel approach to AUC maximization, based on sampling mini-batches of positive/negative instance pairs and computing U-statistics to approximate a global risk minimization problem. The resulting algorithm is simple, fast, and learning-rate free. We show that the number of samples required for good performance is independent of the number of pairs available, which is a quadratic function of the positive and negative instances. Extensive experiments show the practical utility of the proposed method.