High-Level CA Rule Extraction

Richards, Meyer and Packard [richa90] have suggested a way to extract two-dimensional cellular automaton rules directly from experimental data. The same idea, outlined below, can in principle be used in more general contexts. 

Richards, et. al. s idea is to use a genetic algorithm to search through a space of a certain class of cellular automata rules for a local rule that best reproduces the observed behavior of the data. Their learning algorithm (which was applied specifically to sequential patterns of dendrites formed by NH4 Br as it solidifies from a supersaturated solution) starts with no a-priori knowledge about the physical system. R, instead, builds increasingly sophisticated models that reproduce the observed behavior. 

Richards, et. al. comment that while the exact relationship between the rule found by their genetic algorithm and the fundamental equations of motion for the solidification remains unknown, it may still be possible to connect certain features of the learned rule to phenomenological models. 

---