Differential Evolution on a GPGPU: The Influence of Parameters on Speedup and the Quality of Solutions

One challenge in studying the speedup performance of evolutionary optimization techniques, particularly in differential evolution, is that many parameters including crossover rate, F, dimensionality, population size and the complexity of the objective function play an important role. In fact, these same parameters also effect the quality of the obtained results. Therefore, it is important to understand the interaction between these parameters in order to make good choices for these key parameters that drive both the quality and speedup metrics. Thus, the purpose of this paper is to show how parameters such as crossover rate, F, dimension, population size, and calls to evaluation functions can influence the speedup and the quality of solutions in a differential evolution algorithm in high dimension problems. The evaluation was done using a 2^k factorial analysis considering a Schwefel Benchmark Function in a Mat lab implementation running on a general purpose GPU. Results have shown that a reasonable speedup can be reached taking into account a high level of programming, i.e., There are a good trade-off between the required effort to program on GPU in Mat lab and the reached Speedup. On the other hand, results in terms of quality of solutions showed that CPU tends to produce better outcomes in some configurations.