GENETIC PROGRAMMING : THEORY AND PRACTICE

With the recent examples of the human-competitiveness of evolutionary d esign systems, of interest is scaling them up to producing more sophisticated des igns. Here we argue that for computer-automated design systems to scale to pr oducing more sophisticated results they must be able to produce designs with gre a er structure and organization. By structure and organization we mean the ch aracteristics of modularity, reuse and hierarchy (MR&H), characteristics that are found both in man-made and natural designs. We claim that these charac teristics are enabled by implementing the attributes of combination, control-flow and abstraction in the representation and define metrics for measuring MR &H as well as define two measures of overall structure and organization by co mbining the measures of MR&H. To demonstrate the merit of our complexity measu r s, we use an evolutionary algorithm to evolve solutions to different sizes of a ta ble design problem and compare the structure and organization scores of th e best tables against existing complexity measures. We find that our measures b ett r correlate with the complexity of good designs than do others, which suppor ts our claim that MR&H are important components of complexity. We also compare evolution using five representations with different combinations of MR&H e nabled and find that the best designs are achieved when all three of these a tributes are present. The results of this second set of experiments demonstrate that implementing representations with MR&H can greatly improve search performa nce.