Improving the accuracy and robustness of genetic programming through expression simplification

  • Authors:

  • Dale C. Hooper;Nicholas S. Flann

  • Affiliations:

  • Space Dynamics Laboratory, Logan, UT;Utah State University, Logan, UT

  • Venue:
  • GECCO '96 Proceedings of the 1st annual conference on Genetic and evolutionary computation
  • Year:
  • 1996

Quantified Score

Hi-index 0.00

Visualization

Abstract

Genetic Programming (GP) is clearly an inductive learning approach because the program discovered must reproduce the behavior of the desired program over all the input space, not just the space represented in the training examples [Koz92e], [Alt94]. When GP uses only consistency with the training examples as a guide, very large (bloated) programs can result that "overfit" the training examples and therefore perform poorly over the complete input space. To avoid this problem, the search process can be biased by applying Occam's razor to prefer simpler, smaller programs that are more likely to reproduce the desired program. This paper introduces a general way of doing this through expression simplification.