Improving efficiency of multi-kernel learning for support vector machines

Support vector machines (SVMs) have been successfully applied to classification problems. Practical issues Involve how to determine the right type and suitable hyperparameters of kernel functions. Recently, multiple-kernel learning (MKL) algorithms are developed to handle these issues by combining different kernels. The weight with each kernel in the combination is obtained through learning. One of the most popular methods is to learn the weights with semidefinite programming (SDP). However, the amount of time and space required by this method is demanding. In this study, we reformulate the SDP problem to reduce the time and space requirements. Strategies for reducing the search space in solving the SDP problem are introduced. Experimental results obtained from running on synthetic datasets and benchmark datasets of UCI and Statlog show that the proposed approach improves the efficiency of the SDP method without degrading the performance.