Cellular Automata Pattern Recognition and Rule Evolution Through a Neuro-Genetic Approach
Stefania Bandini, Leonardo Vanneschi, Andrew Wuensche and Alessandro Bahgat Shehata
Cellular Automata rules often produce spatial patterns which make them recognizable by human observers. Nevertheless, it is generally difficult, if not impossible, to identify the characteristic(s) that make a rule produce a particular pattern. Discovering rules that produce spatial patterns that a human being would find “similar” to another given pattern is a very important task, given its numerous possible applications in many complex systems models. In this paper, we propose a general framework to accomplish this task, based on a combination of Machine Learning strategies including Genetic Algorithms and Artificial Neural Networks. This framework is tested on a 3-values, 6-neighbors, k-totalistic cellular automata rule called the “burning paper” rule. Results are encouraging and should pave the way for the use of our framework in real-life complex systems models.
Keywords: spatial patterns, pattern recognition, rule evolution, machine learning, hybrid learning systems, neural networks, genetic algorithms.