MOSES (Meta-Optimizing Semantic Evolutionary Search) is a program-evolution algorithm that learns compact, human-readable programs via estimation-of-distribution genetic search.