An Empirical Study of Graph Grammar Evolution

摘要: Finding an optimal topology for a graph is relevant to many problem domains, as graphs can be used model variety of systems. Evolutionary algorithms (EAs) constitute popular class heuristic optimization algorithms, but have mainly been applied what constitutes just small subset graphs, namely string and trees. Methods evolving typically involve the interpretation or tree into (e.g. Shirakawa et al., 2007). Accordingly, they rely on classical variation operators that are proven easy implement, were fundamentally never designed may struggle with their intrinsically greater complexity. Yet operating directly does not necessarily address this either. What needed representation facilitates discovery reuse design dependencies within graphs. Graph grammars key this, application evolutionary building will focus chapter. Grammars performed two distinct roles in context comptuation: (1) means establishing search bias, both declarative preferential, which restrict guide process, respectively; (2) scalable separates complexity genotype from phenotype. Both these eminently useful capabilities rarely found conjunction. We therefore start by reviewing past research trends fields then describe technique Shared Grammar Evolution (SGE), synergistically combines one coherent framework. SGE subsequently evolve Cellular Grammar, tailored change. experimentally explore impact diversity spatial separation convergence, propose new inspired swarm intelligence. Finally, issue bloat efficacy representational analysed so provide practical insight unique scheme.