We study the dynamical behavior of complex adaptive automata during unsupervised learning of periodic training sets. A new technique for their analysis is presented and applied to an adaptive network with distributed memory.