Benchmark Problems on GPU: Accelerating Experiments on Metaheuristics
Leonardo da Luz Dorneles, Mateus Boiani, Márcio Dorn · 2023
Effectiveness analysis on population-based meta-heuristics can be a longstanding task, specifically when dealing with large populations. This makes this class of algorithms computationally expensive. However, its intrinsic parallel characteristic accepts significant reductions in experimentation time using parallel computing, making it possible to obtain results faster and at a lower cost. This work presents a library for speeding up experiments on population-based metaheuristics by computing objective functions on GPUs. Moreover, it provides reliable benchmark functions to test the effectiveness of GPU-based implementations of metaheuristics. The proposed library parallelizes the CEC'22 benchmark problems, and the results demonstrate a significant improvement in experimentation time compared to the original library. With this work, we aim to provide the metaheuristics researcher community with a way to reduce the time required to evaluate these new methods and approaches, allowing the community to focus on more challenging and exciting aspects of the research. We hope this library will be a valuable resource for researchers working on metaheuristics and stimulate the rapid advancement of this important area of study.