Multitask Twin Support Vector Machines

Multitask learning is a learning paradigm which seeks to improve the generalization performance of a task with the help of other tasks. Learning multiple related tasks simultaneously has been empirically as well as theoretically shown to improve performance relative to learning each task independently. In this paper, we propose a new classification method named multitask twin support vector machines based on the regularization principle and twin support vector machines. Our new approach is that we put twin support vector machines to multitask learning. Experimental results demonstrate that the proposed method dramatically improves the performance relative to learning each task independently.