Distributed support vector machines: An overview

As an important tool for data mining, support vector machines (SVMs) have obtained considerable attention in the area of pattern recognition. Recently distributed algorithms especially the distributed support vector machines (DSVMs) are getting increasing attention with the widespread of networks of interconnected devices. In this paper, the state of the art of DSVMs is first reviewed. The idea, advantage and shortcoming of the existing DSVMs including the cascade SVMs, incremental SVMs, distributed parallel SVM, consensus-based SVM etc are analysed. Then the research problem and some open issues related to the distribution of SVM algorithms are presented. The compressed sensing is pointed out to be promising for future research direction for SVM based distributed learning system especially in energy and bandwidth strictly limited sensor networks.