Evaluation of Seed Throwing Optimization Meta Heuristic in Terms of Performance and Parallelizability

Seed Throwing Optimization is an easy to implement probabilistic metaheuristic for multimodal function optimization with roots in hill climbing and the evolutionary computation like technique Harmony Search. It is a randomized Gradient Ascent with multiple initial states and the possibility to limit exploration to only paths which have shown potential. In this paper, the speed of convergence of Seed Throwing Optimization is compared to Multi-Level Gradient Ascent, Harmony Search, Particle Swarm Optimization and Simulated Annealing. Improvements of the Seed Throwing Optimization are presented. Parallelizability of the mentioned metaheuristics is examined. Parts which are suitable for parallelization are extracted by identifying data and control flow dependencies. Two applications, port optimization in minimally invasive surgery and network parameter optimization for a distributed robotic system, are shown. The presented methaheuristics are tested in a benchmark. The highest convergence speed could be achieved with Harmony Search and Seed Throwing Optimization. Keywords-Multimodal function optimization; Parallel computing; Port optimization in minimally invasive surgery; Network optimization.