A novel method to train support vector machines for solving quadratic programming task
Qian Zhang, Zhanbin Che · 2008
Support vector machine (SVM) plays an important role in the data mining and knowledge discovery by constructing a non-linear optimal classifier. The key problem of training support vector machines is how to solve quadratic programming problem, which results in calculation difficulty while learning samples gets larger. The intelligent search techniques, such as genetic algorithm and particle swarm optimization algorithm, can reach a similar solution of problem in less time. In this paper, quantum particle swarm optimization (QPSO) with characteristics of a fast convergence and better stability than the traditional evolutionary algorithms is developed on the basis of classical particle swarm optimization. Both the QPSO and classical algorithm are used to train support vector machines to solve quadratic programming problem. Simulation results show that it is a feasible and effective way for solving quadratic programming problem with a large scale of training sets.