臺灣博碩士論文加值系統

為了從類神經網路 (Neural Network) 得到正確的結果，必須反覆學習 (Learning) 直到每個輸入都能正確對應到所需要的輸出為止。當需要學習的資料數量龐大時，學習的過程往往需要一段很長的時間。因此，許多增快學習速度的方法便被廣泛的研究與討論。在本篇論文中，我們提出一種將資料分散到多台電腦上平行計算的方法，以縮短學習時間。從實驗中可以發現，使用這種方法計算而得的誤差值與傳統非平行方法的誤差值接近。而且，當學習資料內容較複雜時，可以有效地縮短學習所耗費的時間。
在實作上，為了確保計算平台是一個可靠的平行計算環境，我們使用派翠網路理論 (Petri Net Theory) 做為環境架構的設計、排程與驗證工具。為了達到分散式計算的功能，我們採用 CORBA 技術做為中介環境，並選擇最適合於跨平台環境的 Java 程式語言加以實作。

In order to get the correct results from a neural network, iteratively learning is needed until every input has been correctly mapped to desired output. The learning procedure takes a long time when the training data set is extremely large. Researchers have worked on many different methods, such as adopting parallel computing mechanism, to shorten the training time. In this thesis, we propose a design of distributed neural computation to shorten the learning time. Our experimental results reveal that the mean square error is close to the sequential backpropagation learning mode, and the learning time is shorter when the data is complex.
In design, the Petri net theory is taken as the base model tool while developing the distributing system to make sure the reliability of the parallel computation environment. In implementation, we choose the CORBA technology as the middleware and the Java programming language as the programming language.

Chapter 1 Introduction 1
1.1. Overview 1
1.2. Organization of the Thesis 3
Chapter 2 Basic Concepts 4
2.1 Neural Networks 4
2.1.1. Neuron Model 4
2.1.2. Neural Network Architectures 5
2.1.3. Backpropagation Learning Algorithm 6
2.2. Petri nets 8
2.2.1 Basic Definitions 9
2.2.2 Modeling of Parallelism 13
2.2.3 Analysis of Petri Nets 14
2.2.3.1 Boundedness 14
2.2.3.2 Reachability 15
2.2.3.3 Liveness 15
2.2.3.4 Reduction Technology 15
2.3 CORBA Technology 17
2.3.1 Interface Definition Language 18
2.3.2 Stubs and Skeletons 19
2.3.3 Dynamic Invocation 19
2.3.4 Object Adapter 20
2.3.5 Portable Object Adapter 20
2.3.6 Naming Service 21
2.3.7 Event Service 21
Chapter 3 Implementation of Distributed Backpropagation Modeling with
Petri Nets 23
3.1 The Distributed Backpropagation Algorithm 23
3.2 Modeling with Petri Nets 24
3.3 Mapping the Computation Model to CORBA 32
3.4 Mapping CORBA to Java 32
Chapter 4 Experimental Results and Analysis 34
4.1 Experiment 1 34
4.2 Experiment 2 38
4.3 Analysis of the Experimental Results 39
Chapter 5 Conclusions and Future Researches 41
5.1. Conclusions 41
5.2. Future Researches 41
References 43

References
[1] Martin T. Hagan, Howard B. Demuth, and Mark H. Beale, “Neural Network Design,” PWS Publishing Company, 1995.
[2] Simon Haykiu, “Neural Networks, A Comprehension Foundation,” Macmillan College Publishing Company, 1994.
[3] Hee T. Chung and Gi J. Jeon, “Distributed Modeling and Control of Large Scale Systems Using Neural Networks,” IEEE Proceedings of International Joint Conference on Neural Networks, 1993.
[4] James L. Peterson, “Petri Net Theory and the Modeling of Systems,” McGraw — Hill Book Company, 1981.
[5] Alice E. Koniges, San Francisco, and Morgan Kaufmann, “Industrial strength parallel computing,” 2000.
[6] Jamshed N. Patel, Ashfaq A. Khokhar, and Leah H. Jamieson, “Scalability of 2-D Wavelet Transform Algorithms: Analytical and Experimental Results on MPPs,” IEEE Transactions on Signal Processing, Vol.48, No.12, 2000.
[7] OMG, “The Common Object Request Broker: Architecture and Specification Revision 2.4,” 2000.
[8] Bryson, A.E., Jr., and Y.C. Ho, “Applied Optimal Control,” Blaisdell, 1969.
[9] Richard P. Lippmann, “An Introduction to Computing with Neural Nets,” IEEE ASSP Magazine, Vol. 4, April 1987, pp.4-22.
[10] O. L. Mangasarian and M. V. Solodov, “Serial and Parallel Backpropagation Convergence Via Nonmonotone Perturbed Minimization,” OptimizationMethods and Software Vol.4, 1994.
[11] Tadao Murata, “Petri Nets: Properties, Analysis and Applications,” Proceedings of the IEEE, April 1989.
[12] Nikolay Anisimov, Aleksey Kovalenko, and Pavel Postupalski, “Compositional Petri Net Environment,” IEEE Symposium on Emerging Technologies & Factory Automation, 1994.
[13] Edsger Wybe Dijkstra, “Co-operating Sequential Processes,” Programming Languages, Academic Press, 1968.
[14] Genßler, T. and Löwe, W. “Correct Composition of Distributed Systems,” IEEE Proceedings of the 31st TOOLS conference, 1999.
[15] W.M. Zuberek and W.Kubiak, “Timed Petri Net Models of Flexible Manufacturing Cells,” IEEE Proceedings of the 36th Midwest Symposium on Circuits and Systems, Vol.2, 1993, pp.922-925.
[16] W.M. Zuberek and W. Kubiak, “Throughput analysis of manufacturing cells using timed Petri nets,” IEEE International Conference on Systems, Man, and Cybernetics, 1994. Humans, Information and Technology, Vol.2, 1994, pp.1328-1333.
[17] W.M. Zuberek, “Throughput analysis of simple closed timed Petri net models,” IEEE Proceedings of the 36th Midwest Symposium on Circuits and Systems, Vol.2, 1993, pp.930-933.
[18] W.M. Zuberek, “Throughput analysis in timed colored Petri nets,” ISCAS '93, 1993 IEEE International Symposium on Circuits and Systems, Vol.4, 1993, pp.2721-2724.
[19] J. Wang, Y. Deng, and G. Xu, “Reachability analysis of real-time systems using time Petri nets,” IEEE Transactions on Systems, Man and Cybernetics, Part B, Vol.30, 2000, pp.725-736.
[20] H.M. Abbas, M.M. Bayoumi, “On the implementation of backpropagation on the Alex AVX-2 parallel system,” IEEE International Conference on Neural Networks, Vol.2, 1997.