Genetic programming with a norm-referenced fitness function

Geng Li, Xiao Jun Zeng

    Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


    In this paper, we develop a new fitness function based on adjustment of the original fitness function using population performance. We call this new fitness function norm-referenced fitness function since it is motivated by the idea of norm-referenced test. Experiments performed in two benchmarkproblems show that, the norm-referenced fitness function developed is capable of improving the overall performance of GP system. Further analysis of the fitness function reveals that the original fitness function suffers from an implicit bias we named as implicit bias towards exploitation in later generations. This implicit bias pushes the population towards convergence. The norm-referenced fitness developed however does not inherit this bias, and we thinkthis is the main reason why the norm-referenced fitness function is able to outperform the original fitness function. We further study the selection of the newly introduced parameter λ in norm-referenced fitness function and give a number of advices to select the optimal value of the parameter. Copyright 2011 ACM.
    Original languageEnglish
    Title of host publicationGenetic and Evolutionary Computation Conference, GECCO'11|Genet. Evol. Comput. Conf., GECCO
    Number of pages7
    Publication statusPublished - 2011
    Event13th Annual Genetic and Evolutionary Computation Conference, GECCO'11 - Dublin
    Duration: 1 Jul 2011 → …


    Conference13th Annual Genetic and Evolutionary Computation Conference, GECCO'11
    Period1/07/11 → …


    • Fitness function adjustment
    • Internal fitness measure
    • Population performance


    Dive into the research topics of 'Genetic programming with a norm-referenced fitness function'. Together they form a unique fingerprint.

    Cite this