Parallelization of Hooke-Jeeves pattern recognition algorithm by using CUDA for GPGPU

Hooke-Jeeves is an algorithm for solving optimization problems that does not require any information about the gradient of the problem. Algorithm search for solution around the initial point at search space, and replace with the better point if it is succeed. At this stage two solution candidate are obtained per dimension. Thus, it needs more function evaluations when compared to similar methods, that have an negative effect on evaluation time of algorithm. In this study, to reduce this effect, the algorithm is converted to work on general purpose graphical processing units (GPGPU) and influence on execution time is investigated.