Parallelization of genetic operations that takes building-block linkage into account

摘要

We propose a performance enhancement using parallelization of genetic operations that takes highly fit schemata (building-block) linkages into account. Previously, we used the problem of solving Sudoku puzzles to demonstrate the possibility of shortening processing times through the use of many-core processors for genetic computations. To increase accuracy, we proposed a genetic operation that takes building-block linkages into account. Here, in an evaluation using very difficult problems, we show that the proposed genetic operations are suited to fine-grained parallelization; processing performance increased by approximately 30 % (four times) with finegrained parallel processing of the proposed mutation and crossover methods on Intel Core i5 (NVIDIA GTX5800) compared with non-parallel processing on a CPU. Increasing GPU resources will diminish the conflicts with thread usage in coarse-grained parallelization of individuals and will enable faster processing.