Comparison of High Performance Parallel Implementations of TLBO and Jaya Optimization Methods on Manycore GPU

Abstract

The utilization of optimization algorithms within engineering problems has had a major rise in recent years, which has led to the proliferation of a large number of new algorithms to solve optimization problems. In addition, the emergence of new parallelization techniques applicable to these algorithms to improve their convergence time has made it a subject of study by many authors. Recently, two optimization algorithms have been developed: Teaching-Learning Based Optimization and Jaya. One of the main advantages of both algorithms over other optimization methods is that the former do not need to adjust specific parameters for the particular problem to which they are applied. In this paper, the parallel implementations of Teaching-Learning Based Optimization and Jaya are compared. The parallelization of both algorithms is performed using manycore GPU techniques. Different scenarios will be created involving functions frequently applied to the evaluation of optimization algorithms. Results will make it possible to compare both parallel algorithms with regard to the number of iterations and the time needed to perform them so as to obtain a predefined error level. The GPU resources occupation in each case will also be analyzed.