A fast coreset minimum enclosing ball kernel machines

A fast coreset minimum enclosing ball kernel algorithm was proposed. First, it transfers the kernel methods to a center-constrained minimum enclosing ball problem, and subsequently it trains the kernel methods using the proposed MEB algorithm, and the primal variables of the kernel methods are recovered via KKT conditions. Then, detailed theoretical analysis and rigid proofs of our new algorithm are given. After that, experiments are investigated via using several typical classification datasets from UCI machine learning benchmark datasets. Moreover, performances compared with standard support vector machines are seriously considered. It is concluded that our proposed algorithm owns comparable even superior performances yet with rather fast converging speed in the experiments studied in this paper. Finally, comments about the existing problems and future development directions are discussed.