Coupled Multiple Kernel Learning for Supervised Classification

keywords: Multiple kernel learning, non-IIDness, coupled kernels, supervised classification
Multiple kernel learning (MKL) has recently received significant attention due to the fact that it is able to automatically fuse information embedded in multiple base kernels and then find a new kernel for classification or regression. In this paper, we propose a coupled multiple kernel learning method for supervised classification (CMKL-C), which comprehensively involves the intra-coupling within each kernel, inter-coupling among different kernels and coupling between target labels and real ones in MKL. Specifically, the intra-coupling controls the class distribution in a kernel space, the inter-coupling captures the co-information of base kernel matrices, and the last type of coupling determines whether the new learned kernel can make a correct decision. Furthermore, we deduce the analytical solutions to solve the CMKL-C optimization problem for highly efficient learning. Experimental results over eight UCI data sets and three bioinformatics data sets demonstrate the superior performance of CMKL-C in terms of the classification accuracy.
reference: Vol. 36, 2017, No. 3, pp. 618–636