05380751

a

05380751 Barbero, \'Alvaro L\'opez, Jorge Dorronsoro, Jos\'e R. A 4-vector MDM algorithm for support vector training. K{\uu}rkov\'a, V\'era (ed.) et al., Artificial neural networks -- ICANN 2008. 18th international conference, Prague, Czech Republic, September 3--6, 2008. Proceedings, Part I. Berlin: Springer (ISBN 978-3-540-87535-2/pbk). Lecture Notes in Computer Science 5163, 315-324 (2008). 2008 Berlin: Springer EN

doi:10.1007/978-3-540-87536-9_33

Summary: While usually SVM training tries to solve the dual of the standard SVM minimization problem, alternative algorithms that solve the Nearest Point Problem (NPP) for the convex hulls of the positive and negative samples have been shown to also provide effective SVM training. They are variants of the Mitchell-Demyanov-Malozemov (MDM) algorithm and although they perform a two vector weight update, they must compute 4 possible updating vectors before deciding which ones to use. In this work we shall propose a 4-vector version of the MDM algorithm that requires substantially less iterations than the previous 2-vector MDM algorithms and leads to faster training if kernel operations are cached.