Abstract: Pattern recognition based on associative pattern memory (APM) and properties of cycles generated by finite cellular automata. APM addresses (e.g., positions in a two dimensional array) represent states. Cycles are repeating sequences of addresses. Each state is mapped to a “randomly” selected region within the input pattern. Each feature extracted from this region determines one of many next states. All next states (one for each feature type) and all sampled regions are assigned to each state randomly upon APM initialization. The process progresses from state to state, sampling regions of the pattern until the state-transition sequence repeats (generates a cycle). Each feature pattern is represented by one cycle, however different cycles can be derived from one pattern depending on the initial state. Some embodiments use a refractory period assuring a minimum cycle length, making it likely that any given pattern yields only one cycle independent of the initial state.