Implementace neuronové sítě na platformě CUDA

Abstract

Bachelor thesis deals with a serial implementation multi-layer perceptron's feedforward neural network and evolution algorithm of particle swarm for its learning and their following parallel implementation on the CUDA platform in C++. The thesis focus on teoretical problematics of neural network, evolutionary algorithms and parallel platform of CUDA. Then follows the description and design of both implementations. At the end of the thesis individual tests of neural network on XOR problem, approximation sine function and sample data are performed.