ASSET: APPROXIMATE STOCHASTIC SUBGRADIENT ESTIMATION TRAINING FOR SUPPORT VECTOR MACHINES
Sangkyun Lee, Stephen J. Wright · 2012
Subgradient methods for SVMs have been successful in solving the primal formulation with linear kernels. The approach is extended here to nonlinear kernels, and the assumption of strong convexity of the objective is dropped, allowing an intercept term to be used in the classifier.