Genetic Programming using the Karva Gene Expression Language on Graphical Processing Units
Computer Science, Massey University, North Shore 102-904, Auckland, New Zealand
Massey University, Computational Science Technical Note CSTN-171, 2013
@inproceedings{CSTN-171,
author={Alwyn V. Husselmann and K. A. Hawick},
title={Genetic Programming using the Karva Gene Expression Language on Graphical Processing Units},
booktitle={Proc. 10th International Conference on Genetic and Evolutionary Methods (GEM’13)},
year={2013},
number={CSTN-171},
pages={GEM2456},
month={22-25 July},
publisher={WorldComp},
institution={Computer Science, Massey University, Auckland, New Zealand},
keywords={karva language; CUDA; genetic programming; gpu; parallel; optimisation},
owner={kahawick},
timestamp={2013.03.19}
}
Genetic Programming (GP) has been employed in many problem domains, and as a result, it has been the subject of much scientific inquiry. The extensive literature body of GP has reported applications in algorithm discovery, image enhancement and cooperative multi-agent systems, as well as many other areas and disciplines, such as agent-based modelling in Geography and Social Science. As models become more complex, further research toward higher efficiency have been warranted. We discuss solutions to large-scale systems which require automatic programming, and present results of a modified data-parallel implementation of GP based on Gene-expression Programming for Graphical Processing Units (GPUs), as well as a modified Santa Fe Ant Trail problem to measure the efficacy of this algorithm. We present results on algorithm convergence as well as timing performance on both GPU and CPU implementations.
June 7, 2013 by hgpu
