2011/11/02 by Sangkyun Lee, Stephen J. Wright, Lee, Sangkyun +1
Computer Science · #Machine Learning and Algorithms #Machine Learning and Data Classification #Machine Learning and ELM
paper · pdf · doi:10.48550/arxiv.1111.0432
Subgradient algorithms for training support vector machines have been quite\nsuccessful for solving large-scale and online learning problems. However, they\nhave been restricted to linear kernels and strongly convex formulations. This\npaper describes efficient subgradient approaches without such limitations. Our\napproaches make use of randomized low-dimensional approximations to nonlinear\nkernels, and minimization of a reduced primal formulation using an algorithm\nbased on robust stochastic approximation, which do not require strong\nconvexity. Experiments illustrate that our approaches produce solutions of\ncomparable prediction accuracy with the solutions acquired from existing SVM\nsolvers, but often in much shorter time. We also suggest efficient prediction\nschemes that depend only on the dimension of kernel approximation, not on the\nnumber of support vectors.\n