Multiple Kernel Learning Based on Cooperative Clustering

In recent years, kernel methods with single kernel had been challenged by the big data because of its heterogeneousness. In order to exploit the advantages of kernel methods, multiple kernel learning was proposed several years ago. However, the time and space complexity of multiple kernel learning increases greatly due to great amount computation of multiple kernels. So far, the research on improving training efficiency for multiple kernel learning mostly focuses on reducing the complexity of solving the objective function, other than reducing the training set. In this paper, the time performance of multiple kernel learning is improved through shrinking the training sets by introducing cooperative clustering, which is a novel method based on k-means clustering. Applying cooperative clustering to multiple kernel learning problems is proposed to reduce the number of support vectors, and then reduce the time complexity of multiple kernel learning algorithms. Experimental results show that the new method improves the efficiency of multiple kernel learning greatly with a slight impact on classification accuracy.