A MapReduce based glowworm swarm optimization approach for multimodal functions

In optimization problems, such as highly multimodal functions, many iterations involving complex function evaluations are required. Glowworm Swarm Optimization (GSO) has to be parallelized for such functions when large populations capturing the complete function space, are used. However, large-scale parallel algorithms must communicate efficiently, involve load balancing across all available computer nodes, and resolve parallelization problems such as the failure of nodes. In this paper, we outline how GSO can be modeled based on the MapReduce parallel programming model. We describe MapReduce and present how GSO can be naturally expressed in this model, without having to explicitly handle the parallelization details. We use highly multimodal benchmark functions for evaluating our MR-GSO algorithm. Furthermore, we demonstrate that MR-GSO is appropriate for optimizing difficult evaluation functions, and show that high function peak capture rates are achieved. We show with the experiments that adding more nodes would help to solve larger problems without any modifications to the algorithm structure.