Parallelization of automatic classification systems based on support vector machines: Comparison and application to JET database

In learning machines, the larger the training dataset the better model can be obtained. Therefore, the training phase can be very demanding in terms of computational time in mono-processor computers. To overcome this difficulty, codes should be parallelized. This article describes two general purpose parallelization techniques of a classification system based on support vector machines (SVM). Both of them have been applied to the recognition of the L-H confinement regime in JET. This has allowed reducing the training computation time from 70 h to 3 min.