Matrix-Instance-Based One-Pass AUC Optimization

Area under the receiver operating characteristic curve, i.e., AUC, is a widely used performance measure. Traditional off-line and some online AUC optimization methods should store the entire or part of dataset in memory which is infeasible to process big data or streaming data applications. So some scholars develop one-pass AUC optimization (OPAUC) which is independent from the data size. While OPAUC cannot process matrix instances. So we propose a matrix-instance-based one-pass AUC optimization model, i.e., MOPAUC, to overcome such an issue. Related experiments on some benchmark datasets including five image datasets validate that MOPAUC can improve the average AUC, cost little running time with matrix-instance cases. Furthermore, some parameters including regularization parameters and weights have less influence on the average AUC while step sizes have strong influence.