抄録
We present a learning cellular automaton model (LCA) that is an extended generation of the traditional cellular automata, with some self-improving functions. The self-improving learning cellular automaton consists of two parts: the main body and the universal constructor, i.e. a learning cellular automaton can constructs itself through the use of a constructing arm, and it also makes any configurations whose description can be stored on its input tape. We aim at using this model to solve some combinatorial optimization problems in some environments with non-complete prior information.