Using Analytic QP and Sparseness to Speed Training of Support Vector Machines

Abstract

Training a Support Vector Machine (SVM) requires the solution of a very large quadratic programming (QP) problem. This paper proposes an algorithm for training SVMs: Sequential Minimal Optimization,or SMO. SMO breaks the large QP problem into a series of smallest possible QP problems which are analytically solvable. Thus, SMO does not require a numerical QP library. SMO's computation time is dominated by evaluation of the kernel, hence kernel optimizations substantially quicken SMO. For the MNIST database, SMO is 1.7 times as fast as PCG chunking; while for the UCI Adult database and linear SVMs, SMO can be 1500 times faster than the PCG chunking algorithm.