f8905bd3df64ace64a68e154ba72f24c-AuthorFeedback.pdf

In fact, training MU-SVM in primal space is the same3 as training a single layered network architecture and is faster than UP. In addition, Proposition 3 enables utilizing4 state-of-art M-SVM solvers [1,2] for solving MU-SVM. All existing parallel/distributed techniques used for UP also applies for MU-SVM in its primal form. Distributed7 implementation for the dual form of MU-SVM can be achieved through eq. 8 in [3]. This is an ongoing effort.