Paralelizace neuronových sítí

Abstract

This thesis deals with parallelization of Kohonen self-organized maps within the MPI.NET environment. The thesis describes an introduction to artificial neural networks and in detail explains self-organized maps. General principles of parallelism using the message passing model, the Message Passing Interface standard and its implementation MPI.NET are also discussed. Various approaches to the parallelization of Kohonen self-organized maps are described and compared to each other. Several approaches have been implemented in MPI.NET, tested on Windows HPC Server and compared to each other in terms of performance.