Advances in Genetic Programming: Volume 2 PART I

Advances in Genetic Programming: Volume 2 MIT Press ISBN 0-262-01158-1, 538 pp. $50.00, hardbound. MIT Press 1996 Fall catalog, order code ANGAH2, now available.

edited by Peter J. Angeline Kenneth E. Kinnear, Jr.

CONTENTS
Contributors ix
Preface xiii
Acknowledgments xv
1 Genetic Programming's Continued Evolution 1
Peter J. Angeline
I VARIATIONS ON THE GENETIC PROGRAMMING THEME 21
2 A Comparative Analysis of Genetic Programming 23
Una-May O'Reilly, Franz Oppacher

In order to analyze Genetic Programming (GP), this chapter compares it with two alternative adaptive search algorithms, Simulated Annealing (SA) and Stochastic Iterated Hill Climbing (SIHC). SIHC and SA are used to solve program discovery problems posed in the style of GP. In separate versions they employ either GP's crossover operator or a mutation operator. The comparisons in terms of likelihood of success and efficiency show them to be effective. Based upon their success, hybrid versions of GP and hill climbing are designed that improve upon a canonical version of GP. Program discovery practitioners may find it useful to coherently view all the algorithms this chapter considers by using the perspective of evolution.

3 Evolving Programmers: The Co-evolution of Intelligent Recombination Operators 45
Astro Teller

The genetic programming process searches over a fitness landscape. The shape of this landscape is determined by the task to be solved and the representation in which the population members are expressed. The movement through this space is determined by the operators that act to recombine the population members. These factors make it imperative that our search for increased power and understanding in genetic programming include the study and improvement of representations and operators. This chapter describes a process for learning SMART recombination programs in a co-evolutionary process and a new representation for the evolution of algorithms. How these SMART operator programs are created, how they act, how they co-evolve with a main population of programs, and experimental results on their use are the subjects of this chapter.

4 Extending Genetic Programming with Recombinative Guidance 69
Hitoshi Iba, Hugo de Garis

This chapter introduces a recombinative guidance mechanism for GP (Genetic Programming), and shows the effectiveness of our approach using various experiments. Traditional GP blindly combines subtrees, by applying crossover operations. This blind replacement, in general, can often disrupt beneficial building-blocks in tree structures. Randomly chosen crossover points ignore the semantics of the parent trees. Our goal is to exploit already built structures by adaptive recombination, in which GP recombination is guided by ``S-value'' measures. We present various S-value definitions, and show that the performance depends upon the definition.

5 Two Self-Adaptive Crossover Operators for Genetic Programming89
Peter J. Angeline
6 Explicitly Defined Introns and Destructive Crossover in Genetic Programming 111
Peter Nordin, Frank Francone, Wolfgang Banzhaf

Advances in Genetic Programming: Volume 2 PART II PART IV

W.B.Langdon@cs.bham.ac.uk 12 November 1996