Benchmarking cellular genetic algorithms on the BBOB noiseless testbed.

In this paper we evaluate 2 cellular genetic algorithms (CGAs), a single-population genetic algorithm, and a hill-climber on the Black Box Optimization Benchmarking testbed. CGAs are fine grain parallel genetic algorithms with a spatial structure imposed by embedding individuals in a connected graph. They are popular for their diversity-preserving properties and efficient implementations on parallel architectures. We find that a CGA with a uni-directional ring topology outperforms the canonical CGA that uses a bi-directional grid topology in nearly all cases. Our results also highlight the importance of carefully chosen genetic operators for finding precise solutions to optimization problems.