For accelerating the training speed of support vector machines (SVM), a novel “multi-trifurcate cascading (MTC)” architecture,which held the advantages of fast feedback, high utilization rate of nodes, and more feeding support vectors,was proposed. A parallel algorithm for training SVM was designed based on the MTC architecture, and it was proven to converge to the optimal solution strictly. The experimental results showed that the proposed algorithm obtained very high speedup and efficiency, and needed significantly less training time than the Cascade SVM algorithm.
