Multi-task nonparallel support vector machine for classification

Direct multi-task twin support vector machine (DMTSVM) explores the shared information between multiple correlated tasks, then it produces better generalization performance. However, contains matrix inversion operation when solving dual problems, so costs much running time. Moreover, kernel trick cannot be directly utilized in nonlinear case. To effectively avoid above a novel nonparallel (MTNPSVM) including linear and cases is proposed this paper. By introducing ε -insensitive loss instead of square DMTSVM, MTNPSVM avoids takes full advantage trick. Theoretical implication model further discussed. improve computational efficiency, alternating direction method multipliers (ADMM) employed problem. The complexity convergence algorithm are provided. In addition, property sensitivity parameter explored. experimental results on fifteen benchmark datasets twelve image demonstrate validity comparison with state-of-the-art algorithms. Finally, applied to real Chinese Wine dataset, also verifies its effectiveness. • A proposed, which an extension NPSVM for settings. inverse follows SRM principle. ADMM MTNPSVM, greatly improves efficiency solving.