Nested sequential minimal optimization for support vector machines

Abstract

We propose in this work a nested version of the well-known Sequential Minimal Optimization (SMO) algorithm, able to contemplate working sets of larger cardinality for solving Support Vector Machine (SVM) learning problems. Contrary to several other proposals in literature, neither new procedures nor numerical QP optimizations must be implemented, since our proposal exploits the conventional SMO method in its core. Preliminary tests on benchmarking datasets allow to demonstrate the effectiveness of the presented method. © 2012 Springer-Verlag.