Nevroevolucijski algoritem NEAT na grafičnih karticah

magistrsko delo

Blaž Sitar (Author), Aleš Holobar (Mentor)

Abstract

V magistrski nalogi naslavljamo problem implementacije algoritma NeuroEvolution of Augmenting Topologies (NEAT) za delovanje na grafičnih karticah. Algoritem NEAT je genetski algoritem za učenje razvijajočih nevronskih mrež. Izhaja iz področja nevroevolucije, ki v umetni inteligenci uporablja genetske algoritme za generiranje in učenje nevronskih mrež. Algoritem za svoje delovanje porabi veliko strojnih in časovnih virov, zato je implementacija na grafičnih karticah smiselna. Implementacijo smo izvedli v arhitekturi CUDA, ki jo podpirajo grafične kartice podjetja NVIDIA. Hitrost in uspešnost algoritma smo izmerili na petih različnih grafičnih karticah in jo primerjali s hitrostjo in uspešnostjo originalnega algoritma. Ugotovili smo, da je naša implementacija algoritma zadovoljiva, saj je hitrejša in prav toliko uspešna kot originalna implementacija algoritma NEAT.

Keywords

nevroevolucija;NEAT;nevronska mreža;genetski algoritem;magistrske naloge;

Data

Language:	Slovenian
Year of publishing:	2019
Typology:	2.09 - Master's Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	[B. Sitar]
UDC:	004.8.021(043.2)
COBISS:	22891030
Views:	687
Downloads:	111
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Neuroevolution algorithm neat on graphics cards
Secondary abstract:	We address the problem of NeuroEvolution of Augmenting Topologies (NEAT) algorithm implementation for operating on graphics cards. NEAT is a genetic algorithm for learning and evolving neural networks. It’s a member of the neuroevolution algorithms which use genetic algorithms to learn and evolve neural networks in the field of artificial intelligence. Algorithm uses a lot of resources for its operation and is, therefore, suitable for implementation on graphics cards. We implemented it on a CUDA architecture, which is supported by NVIDIA graphics cards. We measured the speed and performance of the algorithm on five different graphics cards and compared it to the speed and performance of the original algorithm. Our implementation is satisfactory because it is faster than and just as efficient as the original NEAT implementation.
Secondary keywords:	neuroevolution;NEAT;neural network;genetic algorithm;CUDA;
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages:	XI, 65 str.
ID:	11219985