high performance computing on graphics processing units: hgpu.org

Several papers have shown that symbolic regression is suitable for data analysis and prediction in ?nance markets. The Grammatical Evolution (GE) has been successfully applied in solving various tasks including symbolic regression. However, performance of this method can limit the area of possible applications. This paper deals with utilizing mainstream graphics processing unit (GPU) for acceleration of GE solving symbolic regression. With respect to various mentioned constrains, such as PCI-Express and main memory bandwidth bottleneck, we have designed effective mapping of the algorithm to the CUDA framework. Results indicate that for larger number of regression points can our algorithm run 636 or 39 times faster than GEVA library routine or a sequential C code, respectively. As a result, the ordinary GPU, if used properly, can offer interesting performance boost for solution the symbolic regression by the GE.