A Hybrid Parallel Implementation of Model Selection for Support Vector Machines

The Model Selection (MS) is an important part of any statistical analysis, and for Support Vector Machine becomes crucial in order to reach the best performance. However, the MS is a compute intensive and non-convex problem, therefore an efficient parallelization is highly desirable. For this reason, in this work we compare two different approaches in MS parallelization, based on the use of MPI and hybrid MPI+OpenMP composition. Results show a clear and considerable advantage in using the latter solution.