The Effects of Transfer of Global Improvements in Genetic Programming

Koza has shown how Automatically Defined Functions (ADFs) can reduce computational effort in the genetic programming paradigm. In Koza's Automatically Defined Functions, as well as in standard genetic programming, an improvement in a part of a program (an ADF or a main body) can only be transferred to other individuals in the population via crossover. In this article, we consider whether it is a good idea to transfer immediately improvements found by a single individual to other individuals in the population. A system that implements this idea has been proposed and tested for the even-5-parity, even-6-parity, and even-10-parity problems. Results are very encouraging: computational effort is reduced (compared to Koza's ADFs) and the system seems to be less prone to early stagnation. Also, as evolution occurs in separate populations, our approach permits to parallelize genetic programming in another different way.

[1]  David H. Wolpert,et al.  The Existence of A Priori Distinctions Between Learning Algorithms , 1996, Neural Computation.

[2]  Astro Teller,et al.  Turing completeness in the language of genetic programming with indexed memory , 1994, Proceedings of the First IEEE Conference on Evolutionary Computation. IEEE World Congress on Computational Intelligence.

[3]  John R. Koza,et al.  Genetic Programming II , 1992 .

[4]  Ricardo Aler,et al.  Experimentacion en Programacion Genetica Multinivel , 2001, Inteligencia Artif..

[5]  Ricardo Aler,et al.  Immediate Transfer of Global Improvements to All Individuals in a Population Compared to Automatically Defined Functions for the EVEN-5, 6-PARITY Problems , 1998, EuroGP.

[6]  Pedro Isasi Viñuela,et al.  Evolving Heuristics for Planning , 1998, Evolutionary Programming.

[7]  Astro Teller,et al.  PADO: a new learning architecture for object recognition , 1997 .

[8]  John R. Koza,et al.  Genetic programming - on the programming of computers by means of natural selection , 1993, Complex adaptive systems.

[9]  Lee Spector,et al.  Evolving Control Structures with Automatically Defined Macros , 2001 .

[10]  David H. Wolpert,et al.  The Lack of A Priori Distinctions Between Learning Algorithms , 1996, Neural Computation.

[11]  W. Daniel Hillis,et al.  Co-evolving parasites improve simulated evolution as an optimization procedure , 1990 .

[12]  John R. Koza,et al.  Genetic programming 2 - automatic discovery of reusable programs , 1994, Complex Adaptive Systems.

[13]  Marc Schoenauer,et al.  A Dynamic Lattice to Evolve Hierarchically Shared Subroutines , 1998, EuroGP.

[14]  Peter Nordin,et al.  Genetic programming - An Introduction: On the Automatic Evolution of Computer Programs and Its Applications , 1998 .

[16]  Justinian P. Rosea.  Towards Automatic Discovery of Building Blocks in Genetic Programming , 1995 .

[17]  Lee Spector,et al.  Cultural transmission of information in genetic programming , 1996 .

[18]  J. Pollack,et al.  The Evolutionary Induction of Subroutines , 1997 .