Category: Science

Summary: Evolving one-dimensional cellular-automaton rules to solve the classic density-classification problem and testing how close evolution gets to known heuristic baselines.

The density-classification task asks a simple but deceptively hard question: can a local cellular automaton decide whether an initial binary string contains more zeros or ones and then converge to the corresponding unanimous state? This experiment revisits that benchmark with a genetic algorithm, asking how well evolved radius-three rules solve the task and how their performance compares with majority-style heuristics.

The script evolves 128-bit rule tables for one-dimensional automata on many random initial conditions, using tournament selection, crossover, and mutation to improve classification fitness over generations. It records the best evolved rules, their accuracy, and comparisons with hand-built baseline rules.

That makes the project a search over distributed computation strategies. The broader interest is in whether simple local update rules can evolve surprisingly effective collective decision behavior on a task that has long served as a benchmark for emergent computation.

Method: Genetic-algorithm search over radius-three one-dimensional cellular-automaton rules for the density-classification task on length-149 lattices.

What is measured: Best evolved fitness, generation-by-generation fitness improvement, final rule-table performance, comparison with majority-rule baselines, and task accuracy across initial conditions.