A HIGH PERFORMANCE OPTIMIZATION TECHNIQUE FOR POLE BALANCING PROBLEM

Abstract

High performance computing techniques can be used effectively for solution of the complex scientific problems. Pole balancing problem is a basic benchmark tool of robotic field, which is an important field of Artificial Intelligence research areas. In this study, a solution is developed for pole balancing problem using Artificial Neural Network (ANN) and high performance computation technique. Algorithm, that basis of the Reinforcement Learning method which is used to find the force of pole's balance, is transfered to parallel environment. In Implementation, C is preferred as programming language and Message Passing Interface (MPI) is used for parallel computation technique. Self–Organizing Map (SOM) ANN model's neurons (artificial neural nodes) and their weights are distributed to six processors of a server computer which equipped with each quad core processor (total 24 processors). In this way, performance values are obtained for different number of artificial neural nodes. Success of method based on results is discussed.

Keywords

• Artificial Intelligence, Pole balancing problem, Parallel computing, Artificial neural network, Learning algorithms.

KUTUP DENGELEME PROBLEMi iÇiN YÜKSEK BASARIMLI BiR OPTiMiZASYON TEKNiGi

Öz

Karmasık bilimsel problemlerin etkin olarak çözümlenmesinde yüksek basarımlı hesaplama teknikleri kullanılmaktadır. Kutup dengeleme problemi, yapay zeka alanları içerisinde önemli yer tutan robotbilim dalının en temel ölçüm araçlarından biridir. Bu çalısmada kutup dengeleme problemi, Yapay Sinir Agı (YSA) ve yüksek basarımlı hesaplama teknigi kullanılarak çözülmüstür. Kutbu (çubugu) dengede tutmayı saglayan kuvvetin bulunmasında kullanılan destekleyici ögrenme yöntemini temel alan algoritma paralel ortama aktarılmıstır. Gerçeklestirimde C programlama dili ve paralel hesaplama teknigi için Mesaj Geçme Arayüzü kullanılmıstır. Bir YSA modeli olan Öz-örgütlemeli Harita Agı'na ait yapay sinir hücre dügümleri ve agırlıkları her biri dört çekirdekli altı adet (toplamda yirmi dört) islemciye sahip bir sunucu bilgisayardaki islemcilere dagıtılarak, farklı sinir hücre sayıları için performans degerleri elde edilmistir. Yöntemin basarısı sonuçlar üzerinden tartısılmıstır.

Anahtar Kelimeler

• Yapay zeka, Kutup dengeleme problemi, Paralel hesaplama, Yapay sinir agları, Ögrenme algoritmaları.

Kaynakça

1. Amdahl, G.M. 1967. Validity of single-processor approach to achieving large-scale computing capability, proceedings of AFIPS conference, reston, VA. p. 483-485.
2. El-Rewini, H. and Abd-El-Barr, M. 2005. Advanced Computer Architecture and Parallel Processing, Wiley-Interscience, John Wiley and Sons Inc.
3. Flynn, M. 1972. Some computer organizations and their effectiveness, IEEE Trans. Comput., Vol. C-21, p. 948.
4. Gomez, F. and Miikkulainen, R. 1998. 2-D pole balancing with recurrent evolutionary networks. In Proceedings of the International Conference on Artificial Neural Networks (ICANN-98), Skovde, Sweden), 425–430.
5. Grama, A., Gupta A., Karypis G. and Kumar, V. 2003. Introduction to Parallel Computing, Second Edition, Addison Wesley Publishing, ISBN 0-201- 64865-2, 856, 2003.
6. Grounds, M. and Kudenko, D. 2006. Parallel reinforcement learning by merging function approximations. Department of Computer Science University of York.
7. Karasulu, B. ve Uğur, A. 2007. Özörgütlemeli yapay sinir ağı modelinin kullanıldığı kutup dengeleme problemi için paralel hesaplama tekniği ile bir başarım eniyileştirme yöntemi, akademik bilişim 2007, Bildiri No: 25, Dumlupınar Üniversitesi, Kütahya, 31 Ocak - 2 Şubat 2007.
8. Kohonen, T. 1996. The Speedy Som, Technical Report a33, Helsinki University of Technology, Laboratory of Computer and Information Science, Espoo, Finland.

9. Kohonen, T. 1997. Self-Organizing Maps, Series in Information Sciences, Vol. 30, Second Edition Springer, Heidelberg, ISBN 3-540-62017-6, 1997.
10. Kretchmar, R. M. 2002. Parallel reinforcement learning. In proceedings of the 6th World Conference on Systemics, Cybernetics, and Informatics (SCI2002). LAM-MPI
11. http://www.lam-mpi.org). websitesi.
12. (Çevrimiçi: Stanley, O.K. and Miikkulainen, R. 2002. Efficient reinforcement learning through evolving neural network topologies. In Proceedings of the Genetic and (GECCO-2002).
13. Conference Pardoe, D., Ryoo, M. and Miikkulainen, R. 2005. Evolving neural network ensembles for control problems, in proceedings of the genetic and evolutionary computation conference (GECCO- 2005).
14. Rauber, A., Tomsich, P. and Merkl, D. 2000. parSOM: A parallel implementation of the self- organizing map exploiting cacheeffects: making the som fit for interactive high performance data analysis, p. 6177, IEEE-INNS-ENNS Int. Joint Conference on Neural Networks (IJCNN'00), Vol 6. Sutton, R. S. 1992. Reinforcement learning architectures. Proceedings ISKIT'92 International Symposium on Neural Information Processing, Fukuoka, Japan.
15. Sutton, R. S. and Barto, A. G. 1998. Reinforcement learning: An introduction., Cambridge, MA, MIT Press.
16. Vishwanathan, S. V. N. and Murty, M. N. 2000. Kohonen’s SOM with cache. The Journal of Pattern Recognition Society 33, 1927-1929.