Stofa 257 - Langholt
Master's student: Kveldúlfur Þrastarson
Title: Design, Analysis and Implementation of a Parallel and Scalable Cascade Support Vector Machine Framework
Faculty: Faculty of Industrial Engineering, Mechanical Engineering and Computer Science
Advisors: Morris Riedel, adjunct associated professor at the Faculty of Industrial Engineering, Mechanical Engineering and Computer Science and dr. Helmut Neukirchen, professor at the Faculty of Industrial Engineering, Mechanical Engineering and Computer Science.
Examiner: Dr. Lars Hoffmann, scientist at the Climate Science Simulation Laboratory, Juelich Supercomputing Centre, Germany.
Natural sciences and engineering are experiencing a growing influx of large datasetsthat are analyzed using machine learning methods, both supervised and unsupervised.This in turn calls for scalable solutions that take advantage of parallel computingsystems such as large supercomputing resources. Cascade support vectormachine training offers a substantial boost to performance when processing largemachine learning training data compared to more traditional parallel methods, withminimal loss of accuracy. This paper describes the design and implementation of acascade SVM system that attempts to be highly parallel and provide a reasonablenumber of SVM features. It also provides far better scaling than common parallelmethods. The proposed approach attempts to keep CPU utilization high by mixingcascade and traditional parallel training methods, utilizing otherwise unusedthreads as workers in the more traditional parallel approach as they are freed up bythe cascade method. Results indicate a 2.7 fold speedup in some tested datasets,improved scaling, and over 80% reduced memory footprint compared to traditionalparallel methods.