2019/03/08 by Babak Hosseini, Barbara Hammer, Hosseini, Babak +1
Computer Science · Biochemistry, Genetics and Molecular Biology · #Machine Learning and ELM #Cancer-related molecular mechanisms research #Face and Expression Recognition
paper · pdf · doi:10.48550/arxiv.1903.03364
Multiple kernel learning (MKL) algorithms combine different base kernels to\nobtain a more efficient representation in the feature space. Focusing on\ndiscriminative tasks, MKL has been used successfully for feature selection and\nfinding the significant modalities of the data. In such applications, each base\nkernel represents one dimension of the data or is derived from one specific\ndescriptor. Therefore, MKL finds an optimal weighting scheme for the given\nkernels to increase the classification accuracy. Nevertheless, the majority of\nthe works in this area focus on only binary classification problems or aim for\nlinear separation of the classes in the kernel space, which are not realistic\nassumptions for many real-world problems. In this paper, we propose a novel\nmulti-class MKL framework which improves the state-of-the-art by enhancing the\nlocal separation of the classes in the feature space. Besides, by using a\nsparsity term, our large-margin multiple kernel algorithm (LMMK) performs\ndiscriminative feature selection by aiming to employ a small subset of the base\nkernels. Based on our empirical evaluations on different real-world datasets,\nLMMK provides a competitive classification accuracy compared with the\nstate-of-the-art algorithms in MKL. Additionally, it learns a sparse set of\nnon-zero kernel weights which leads to a more interpretable feature selection\nand representation learning.\n