Research Of Parallel Computing Apply In Support Vector Machine Under MATLAB Cluster

Parallel computing is a new generation of computer science and computational mathematics combination, it brings new solutions to some algorithm facing the bottleneck, through the high performance computing mode, complete a variety of algorithm complex operation part, improving the time efficiency. Support vector machine is a pattern recognition methods based on statistical learning theory, can successfully deal with classification and regression problem, it also has many application direction. However, when face the huge iterative process as well as a large number of samples in the training problem, algorithm efficiency is not satisfactory, the traditional algorithm improvement is hard to meet the requirements of effect. In this paper, the idea of parallel computing is introduced into various areas of applications of support vector machines to solve the aforementioned problems, and the paper’s content is about both directions of the multi-target tracking and stock forecasts.Support vector machine can be used in many field of multiple targets tracking. In Multi-target tracking system, data association is the core part. And in dense clutter environment, especially if the intersection of tracking gates exist too much clutter, filter values of probabilistic data association algorithm will diverge, ultimately system’s track performance will be seriously interfered. In our paper, a novel algorithm called SVM-MC-JPDAF is proposed, which introduces Support Vector Machine (SVM) to classify valid echo in intersection of tracking gates, and we use distributed parallel way to solve the problem that new algorithm’s computational burden become large. Through simulation experiment, the algorithm effectively reduced the loss rate and values of Root Mean Square Error (RMSE), then parallelize the algorithm, improving the performance of tracking simultaneously controlling algorithm implementation burden.Support vector machine applied to financial stock market prediction has also made good results, but with the stock data size and dimensions of sample become larger increasingly, for stock prediction algorithm, we not only require accuracy, but also a higher speed. This paper presents a new GPU parallel computing model to improve least squares support vector machine (LSSVM) stock prediction algorithm’s speed. Experiments show that the new method not only can guarantee the accuracy of prediction, and can greatly reduce the prediction time. This method could be widely applied to finance large-scale data processing and forecasting, with a high value.