Improving the generalisation ability of genetic programming with semantic similarity based crossover

  • Authors:
  • Nguyen Quang Uy;Nguyen Thi Hien;Nguyen Xuan Hoai;Michael O'Neill

  • Affiliations:
  • Natural Computing Research & Applications Group, University College Dublin, Ireland;School of Information Technology, Vietnamese Military Technical Academy, Vietnam;School of Information Technology, Vietnamese Military Technical Academy, Vietnam;Natural Computing Research & Applications Group, University College Dublin, Ireland

  • Venue:
  • EuroGP'10 Proceedings of the 13th European conference on Genetic Programming
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper examines the impact of semantic control on the ability of Genetic Programming (GP) to generalise via a semantic based crossover operator (Semantic Similarity based Crossover - SSC). The use of validation sets is also investigated for both standard crossover and SSC. All GP systems are tested on a number of real-valued symbolic regression problems. The experimental results show that while using validation sets barely improve generalisation ability of GP, by using semantics, the performance of Genetic Programming is enhanced both on training and testing data. Further recorded statistics shows that the size of the evolved solutions by using SSC are often smaller than ones obtained from GP systems that do not use semantics. This can be seen as one of the reasons for the success of SSC in improving the generalisation ability of GP.