We address the problem of learning a non-sparse linear combination of Multiple kernels for Support vector Machines. Existing approaches often emphasize on sparsity which could be sub-optimal in many situations. In this talk we will motivate a general mixed norm regularization for learning Multiple kernels. One can obtain both sparse and non-sparse variants of Multiple Kernel learning formulations by choosing an appropriate norm. We will discuss an provably convergent iterative algorithm, where each iteration is no difficult than solving an SVMs. Experimental results show that the algorithm is scalable, and yields accurate classifiers.
Based on joint works with: Saketha Nath J. , G. Dinesh, Raman S., Aaron Ben-Tal, K. R. Ramakrishnan
BioChiranjib Bhattacharyya is an Associate Prof. in the Dept of CSA, Indian Institute of Science(IISc). His research interests are in Machine Learning.