Learning algorithm for modeling complex spatial dynamics

Abstract
A learning algorithm is developed to build a dynamical model from complex spatial data consisting of discrete values on a lattice evolving in time. The resulting dynamical system is a cellular automaton, and may be used for forecasting or for regenerating global spatial patterns. Part of the learning algorithm is a novel application of the genetic algorithm, originally developed in the field of machine learning. We outline extensions of this method to construct models for spatial dynamics that have continuous variables as well.

This publication has 8 references indexed in Scilit: