Complexity as Fitness for Evolved Cellular Automata Update Rules

We investigate the state change behavior of one-dimensional cellular automata during the solution of the binary density-classification task. Update rules of high, low and unknown fitness are applied to cellular automata, thereby providing examples of high and low rates of successful classification. A spread factor, ω, is introduced and investigated as a numerical marker of state change behavior. The nature of ω describes complex or particle-like behavior on the part of the cellular automata over the middle region of initial configuration density-distribution, but breaks down at the ends. Because of the limitation on ω, a related jump-out term, jot, is selected for incorporation into the finess function for genetic algorithm evolution of update rules. The inclusion of jot in the fitness function significantly reduces the number of generations required to reach high rates of successful classification (≥90%).