Evolving cellular automata to perform computations: mechanisms and impediments

Optimization as Side-Effect of Evolving Allelopathic Diversity

“…a cellular automaton with 100% classification accuracy, does not exist [8]. In previous studies of (co-)evolutionary optimization models that used this optimization problem it appeared difficult to evolve CAs with high performance values [9,10,16]. Recently, however, cellular automata have been found with performance values of up to 0.86 [6].…”

“…The traditional point mutation operator, i.e. flipping a bit at a random position in the string, gives a strong bias towards initial conditions with a density value of 0.5 (see also [10]). Thus, the mutation operator is biased with respect to the property of ICs on the basis of which they are to be classified.…”

“…by flipping as little as a single bit). In fact, the quality of classification of a 'good' cellular automaton, like for instance the GKL rule, decreases rapidly if it is evaluated on the basis of ICs whose density approaches 0.5 [6,10]. The traditional point mutation operator, i.e.…”

“…The GP-mapping that we use in this model is based on a well studied optimization problem, the density classification task for cellular automata [2,6,9,10,15,16]. The genes that code for toxins and antidotes are represented by bit strings.…”

Computation in Cellular Automata: A Selected Review

“…This process was repeated for 100 generations for a single run of the GA. (Since a different sample of ICs was chosen at each generation, the fitness function was stochastic.) For a discussion of this algorithm and details of its implementation, see Mitchell, Crutchfield, and Hraber (1994b).…”