Derivative operators for preference predicate evolution

  • David H. Lorenz
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 866)


This work deals with the problem of function learning by genetic algorithms where the function is used as a preference predicate. In such a case, learning the exact function is not necessary since any function that preserves the order induced by the target function is sufficient. The paper presents a methodology for solving the problem with genetic algorithms. We first consider the representation issues involved in learning such a function, and conclude that canonical representation, relative coding, and search restrictions, are required. We then show that the traditional homologous genetic operators are not appropriate for such learning, and introduce a new configurable analogous genetic operator, named derivative crossover. This operator works on the derivative of the chromosomes and is therefore suitable for preference predicate learning where only the relative values of the functions are important. We support our methodology by a set of experiments performed in the domain of continuous function learning and in the domain of evaluation-function learning for game-playing. The experiments show that indeed using derivative operators increases the speed of learning significantly.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    P. J. Angeline and J. B. Pollack. Competitive envinments evolve better solutions for complex tasks. Proc. of the 5 th International Conference on Genetic Algorithms, pages 264–270, July 17–21, 1993.Google Scholar
  2. 2.
    H. J. Antonisse. A grammer-based genetic algorithm. In G. J. Rawlins, editor, Foundations of Genetic Algorithms. Morgan Kaufmann, 1991.Google Scholar
  3. 3.
    M. F. Bramlette. Initialization, mutation and selction methods in genetic algorithms for function optimization. In Proc. of the Fourth International Conference on Genetic Algorithms, pages 100–107, San Diego, CA, 1991.Google Scholar
  4. 4.
    Y. Davidor. Analogous crossover. In Proc. of the 3 rd International Conference on Genetic Algorithms, pages 98–103. Morgan Kaufman, 1989.Google Scholar
  5. 5.
    B. R. Fox and M. B. McMahon. Genetic operators for sequencing problems. In G. J. Rawlins, editor, Foundations of Genetic Algorithms, pages 284–300. Morgan Kaufmann, 1991.Google Scholar
  6. 6.
    D. E. Goldberg. GENETIC ALGORITHMS in Search, Optimazation and Machine Learning. Addison-Wesley, Reading, MA, 1989.Google Scholar
  7. 7.
    D. H. Lorenz and S. Markovitch. Derivative evaluation function learning using genetic operators. In Games: Planning and Learning, Papers from the 1993 Fall Symposium, Raleigh, North Carolina, Oct. 22–24, 1993. American Association for Artificial Intelligence, AAAI Press, Menlo Park, California. Technical Report FS9302.Google Scholar
  8. 8.
    T. M. Mitchell. Generalization as search. Artificial Intelligence, 18:203–226, 1982.CrossRefGoogle Scholar
  9. 9.
    L. Rendell. Induction as optimization. IEEE Transactions on Systems, Man, and Cybernetics, 20(2):326–338, Mar 1990.Google Scholar
  10. 10.
    P. E. Utgoff and S. Saxena. Learning a preference predicate. In Proceedings of the 4 th International Workshop on Machine Learning, pages 115–121. Morgan Kaufman, 1987.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1994

Authors and Affiliations

  • David H. Lorenz
    • 1
  1. 1.Department of Computer ScienceTechnion - Israel Institute of TechnologyIsrael

Personalised recommendations