Towards an OpenCL Implementation of Genetic Algorithms on GPUs

被引：0

作者：

Puzniakowski, Tadeusz ^{[1
]}

Bednarczyk, Marek A. ^{[2
,3
]}

机构：

[1] Univ Gdansk, Inst Informat, PL-80952 Gdansk, Poland

[2] Polish Japanese Inst Informat Technol, Fac Informat, PL-80952 Gdansk, Poland

[3] Polish Acad Sci, Inst Comp Sci, PL-00901 Gdansk, Poland

来源：

SECURITY AND INTELLIGENT INFORMATION SYSTEMS | 2012年 / 7053卷

关键词：

D O I：

暂无

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

The paper compares usual sequential implementations in C of a Genetic Algorithm with parallel implementations in OpenCL. It turns out that the speedup obtained by turning parallel depends on the choice of the selection methods used in CA. In particular the simple tournament selection method yields better results than the selection based on the roulette rule. In case of the latter which requires a synchronization of threads which manipulate individual chromosomes. This is done to compute the joint fitness of a population and find the best specimen. With the help of scan method this can be achieved with O(log n) complexity.

引用

页码：190 / +

页数：3