Parallel classification using support vector machines on Mmodern GPUs

Abstract

The objective of this thesis will be the design, development and evaluation of an efficient parallel SVM classifier in a CUDA programming environment. The development of the above classifier will be done using C/C++, while the evaluation will be performed in a real, modern environment using modern GPUs (NVIDIA 1060 και NVIDIA RTX TITAN) and will include comparisons (latency, speedup and accuracy of classification) with an equivalent sequential implementation.