Experiments Assessing Learning of Agent Behavior using Genetic Programming with Multiple Trees

Takashi Ito, Kenichi Takahashi, Michimasa Inaba

2014

Abstract

In this paper, experiments to assess agent behavior learning are conducted to demonstrate the performance of genetic programming (GP) with multiple trees. Using the methods, each has a chromosome representing agent behavior as several trees. We have proposed two variants using the conditional probability and the island model to improve the methods’ performance. In GP using the conditional probability, individuals with high fitness values are used to produce conditional probability tables to generate individuals in the next generation. In GP using the island model, the population is divided into two islands of individuals: one island maintains diversity of individuals. The other emphasizes the accuracy of the solution. Moreover, this paper improves methods to seek the optimal number of executions of each tree in an individual. Those methods are applied to a garbage collection problem and a Santa Fe Trail problem. They are compared with traditional GP, GP with control nodes, and genetic network programming (GNP) with control nodes. Experimental results show that our methods are effective for improving the fitness.

References

  1. Koza, J. R., 1992. Genetic Programming: On the Programming of Computers by Natural Selection, Cambridge, MA: MIT Press.
  2. Hirasawa, K., Okubo, M., Katagiri, H., Hu, J., and Murata, J., 2001. Comparison between Genetic Network Programming and Genetic Programming Using Evolution of Ant's Behaviors, IEEJ Transactions on Electronics, Information and System, Vol.121, No.6, pp.1001-1009.
  3. Iba, H., 2002. Genetic Algorithm, Igaku Shuppan. Japan.
  4. Mesot, B., Sanchez, E., Pena, C.-A., and Perez-Uribe, A., 2002. SOS++: Finding Smart Behaviors Using Learning and Evolution, Eighth International Conference on the Simulation and Synthesis of Living Systems (Alife 8), Artificial Life 8, pp.264-273.
  5. Tanji, M., and Iba, H., 2010. A New GP Recombination Method Using Random Tree Sampling, IEEJ Transactions on Electronics, Information and Systems, Vol.130, No.5, pp.775-781.
  6. Iba, H., 2002. Genetic Programming, University of Tokyo Press.
  7. Murata, T., and Nakamura, T., 2006. Multi-Start Node Genetic Network Programming for Controlling Multiple Agents, 2006 IEEE International Conference on Systems, Man, and Cybernetics, Vol.3, pp.1927- 1932.
  8. Eto, S., Mabu, S., Hirasawa, K., Huruzuki, T., 2007. Genetic Network Programming with Control Nodes, 2007 IEEE Congress on Evolutionary Computation (CEC2007), pp.1023-1028.
  9. Minesaki, T., Ueda, H., and Takahashi, K., 2009. Comparison experiment using Genetic Network Programming, The Conference Program of the 2009 (60th) Chugoku-branch Joint Convention of Institutes of Electrical and Information Engineers, p.546.
  10. Morioka, T., Ueda, H., and Takahashi, K., 2011. Efficient Evolutionary Learning of Agent Behavior by Genetic Programming Using the Conditional Probabilities, Proc. of 12th International Symposium on Advanced Intelligent System 2011 (ISIS2011), pp.342-345.
  11. Ito, T., Takahashi, K., Inaba, M., 2013. Improvement of Genetic Programming with multiple trees, 2013 IEEE SMC Hiroshima Chapter Young Researchers' Workshop Proceedings, pp.9-12.
  12. Ono, K., Hanada, Y., Shirakawa, K., Kumano, M., Kimura, M., 2012. Depth-dependent crossover in genetic programming with frequent trees, 2012 IEEE International Conference on Systems, Man, and Cybernetics, pp.359-363.
  13. Ono, K., Hanada, Y., Kumano, M., Kimura, M., 2013. Island model genetic programming based on frequent trees, 2013 IEEE Conference on Evolutionary Computation (CEC2013), pp.2988-2995.
  14. Iwashita, M., and Iba, H., 2002. Parallel Distributed GP with Immigrants Aging and Depth-dependent Crossover, Transactions of Information Processing Society of Japan, Vol.43, No.SIG10, pp.146-156.
Download


Paper Citation


in Harvard Style

Ito T., Takahashi K. and Inaba M. (2014). Experiments Assessing Learning of Agent Behavior using Genetic Programming with Multiple Trees . In Proceedings of the 6th International Conference on Agents and Artificial Intelligence - Volume 1: ICAART, ISBN 978-989-758-015-4, pages 264-271. DOI: 10.5220/0004751402640271


in Bibtex Style

@conference{icaart14,
author={Takashi Ito and Kenichi Takahashi and Michimasa Inaba},
title={Experiments Assessing Learning of Agent Behavior using Genetic Programming with Multiple Trees},
booktitle={Proceedings of the 6th International Conference on Agents and Artificial Intelligence - Volume 1: ICAART,},
year={2014},
pages={264-271},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0004751402640271},
isbn={978-989-758-015-4},
}


in EndNote Style

TY - CONF
JO - Proceedings of the 6th International Conference on Agents and Artificial Intelligence - Volume 1: ICAART,
TI - Experiments Assessing Learning of Agent Behavior using Genetic Programming with Multiple Trees
SN - 978-989-758-015-4
AU - Ito T.
AU - Takahashi K.
AU - Inaba M.
PY - 2014
SP - 264
EP - 271
DO - 10.5220/0004751402640271