Learning Cellular Automata for Function Optimization Problems

抄録

We present a model of learning cellular automata (LCA) as an emergent system having some collective behaviors. LCA is an extended version of the traditional cellular automaton. Especially, we adopt the LCA with some self-improving functions, called self-improving learning cellular automata (SILCA) and develop its optimization capability. Each self-improving learning cellular automaton, i.e. a member of SILCA, consists of two parts: main body and a universal constructor. Through the use of a constructing arm, a learning cellular automaton is capable of constructing any configuration whose description can be stored on its input tape. As an example of combinatorial optimization problems, we consider function optimization problems and show the SILCA's emergent capability for optimization.