Evolving weighted topologies for neural networks using genetic programming

• Autorzy: Suchorzewski M.
• Języki publikacji: EN

Abstrakty

EN

NeuroEvolution of Augmenting Topologies (NEAT) is a method of evolving weighted topologies for neural networks, which has been showed to be effective in solving double pole balancing problems. Its usability, however, is largely reduced by its high specificity and custom architecture. In this paper we propose much more standardized framework for evolving neural networks, achieving comparable or even better performance in three benchmark problems. We show that tree-based genetic programming (GP) requires only minor modifications to represent and evolve neural networks. Despite its high performance, however, we remain sceptical about the robustness of the proposed method. In a series of further experiments we demonstrate the performance is sensitive to parameter settings and algorithm details. Therefore we refrain from making conclusions about performance of our method and only stress the improvement regarding its simplicity.

Słowa kluczowe

EN

NeuroEvolution of Augmenting Topologies; NEAT; genetic programming; neural networks; parameter setting

• Wydawca: Komisja Informatyki Polskiej Akademii Nauk, Oddział w Gdańsku
• Czasopismo: Metody Informatyki Stosowanej
• Rocznik: 2008
• Tom: nr 4 (Tom 17)
• Strony: 211--219
• Opis fizyczny: Bibliogr. 12 poz., rys.

Twórcy

autor

Suchorzewski M.

• Szczecin University of Technology, Faculty of Computer Science and Information Technology

Bibliografia

• [1] Bartz-Beielstein T., Preuss M. Experimental research in evolutionary computation. In Proceedings of the 2007 GECCO conference companion on Genetic and evolutionary computation, pages 3001-3020. ACM Press New York, NY, USA, 2007.
• [2] Daida J., Ampy D., Ratanasavetavadhana M., Li H., Chaudhri O. Challenges with verification, repeatability, and meaningful comparison in genetic programming: Gibsons magic. In Proceedings of the Genetic and Evolutionary Computation Conference, volume 2, pages 1851-1858, 1999.
• [3] Eiben A. E., Michalewicz Z., Schoenauer M., Smith J. E.. Parameter control in evolutionary algorithms. In Parameter Setting in Evolutionary Algorithms, Studies in Computational Intelligence. Springer, 2007.
• [4] James D., Tucker P. A comparative analysis of simplification and complexification in the evolution of neural network topologies. In Keijzer M. (ed.), Late Breaking Papers at the 2004 Genetic and Evolutionary Computation Conference, Seattle, Washington, USA, 26 July 2004.
• [5] Koza J. R. Genetic Programming. MIT Press, 1992.
• [6] Poli R. Parallel distributed genetic programming. In Corne D., Dorigo M., Glover F. (eds.), New Ideas in Optimisation, chapter 27, pages 403-432. McGraw-Hill Ltd., Maidenhead, UK, 1999.
• [7] Poli R., Langdon W., McPhee N. A Field Guide to Genetic Programming. Lulu Press, 2008.
• [8] Pujol J., Poli R.. Evolving the topology and the weights of neural networks using a dual representation. Applied Intelligence, 8(1):73-84, 1998.
• [9] Stanley K., Kohl N., Sherony R., Miikkulainen R. Neuroevolution of an automobile crash warning system. In GECCO’05: Proceedings of the 2005 Conference on Genetic and Evolutionary Computation, pages 1977-1984, New York, NY, USA, 2005. ACM.
• [10] Stanley K. O., Bryant B. D., Miikkulainen R. Real-time evolution in the NERO video game. In Proceedings of the IEEE 2005 Symposium on Computational Intelligence and Games. IEEE, 2005.
• [11] Stanley K. O., Miikkulainen R. Evolving neural networks through augmenting topologies. Evolutionary Computation, 10(2):99-127, 2002.
• [12] Stanley K. O., Miikkulainen R. Competitive coevolution through evolutionary complexification. Journal of Artificial Intelligence Research, 21:63-100, 2004.