大部的類神經網路均為解決特定問題而設計，並非真正去模擬人腦的功能

，在本論文中介紹一個模擬人類學習方式的類神經網路，稱為推理類神經

網路（The Reasoning Neural Network），其主要兩個組成為強記（

cram -ming）及推理（reasoning）部份，透過彈性的組合這兩個部份可

使類神經網路具有類似人類的學習程序。在本論文中介紹其中一個學習程

序並用四個實驗來評估推理類神經網路的績效，從實結果得知，推理類神

經網路能以合理的隱藏節點數（hidden nodes）達到學習的目標，並建立

一個網路內部表示方式（internal representation），及具有好的推理

能力（g eneralization ability）。

Most of artification Neural Networks are designed to resolve

spe -cific problems, rather than to model the brain. The

Reasoning N -eural Network (RNN) that imitates the way of human

learning is presented here. Two key components of RNN are the

cramming and t -he reasoning. These components coulds be

arranged flexibly to a -chieve the human-like learning

procedure. One edition of the RNN used in experiments is

introduces, and four different proble -ms are used to evaluate

the RNN's performance. From simulation results, the RNN

accomplishes the goal of learning with a reason -able number of

hidden nodes, and evolves a good internal repres -entation and

a generalization ability.

Bibliography
中文部份
[葉怡成93]葉怡成，類神經網路模式應用與實作，儒林圖書公司，民國八十三年.
英文部份

[Arai89] Arai M., "Mapping abilities of three-Layer Neural Networks, ", In
Proceedings of the International Conference on Neural Networks
(Washington, DC:IEEE), pp I:419-424, 1989.
[Arai93] Arai M., "Bounds on the Number of Hidden Nodes in Binary-Valued
Three-Layer Neural Networks,", In IEEE Transactions on Neural
Networks vol 6, pp 855-860, 1993.
[Chau89] Y. Chauvin, "A back-propagation algorithm with optimal use of hidden
node," In Advance in Neural Information Processing (2), D.S . Toutetzky,
Ed. (Denver 1988), 1989,pp. 519-526.
[Cybe89] G. Cybenko, "Approximation by superpositions of a sigmoidal function,"
Mathematics of Control, Signals, and Systems, 2(4): pp. 303-314, 1989.
[Cast93] Castellano G., Fanelli A. M., Pelillo M. ,"An Empirical Comparsion of
Node Pruning Methods for Layerd Feed-forward Neural Networks", In
Proceedings of 1993 International Joint Conference on Neural Networks,
pp 321-326, 1993.
[Cun90] Y. Ie Cun, 1.s. Denker, and S.A. Solla, "Optimal brain damage." In D.
Tourezky, editor, Advances in Neural Information ProceSSing System 2, pp
598-605. Morgan Kaufmann, 1990.
[Free92] Freeman 1. A. , Skapura D. M.,"Neural Networks Algorithm, Applications,
and Programming Techniques", Addison-Wesley, Reading MA, 1992.
[Hertz91] J. Hertz, A. Krogh, R. Palmer, "Introduction to the theory of neural
computation", Addison-Wesley, Reading MA, 1991.
[Hagi93] Hagiwara M.,"Removal of hidden nodes and Weights for Back
Propagation Networks", In Proceedings of 1993 International Joint
Conference on Neural Networks, pp 35 1-353,1993.
[Hajn87] A Hajnal, W. Maass, P.Pudlak, M. Szegedy, and G. Turan, "Threshold
circuits of bounded depth." In Proceedings of the J 987 IEEE Symposium
on the Funciations of Science, pp. 99-110,1987.
[Hint87] Geoffrey E. Hinton and Terrence J. Sejnomsk.i. "Neural network
architectures for AI" Tutolial No. MJ?2, AAA187, Seattle, W A, July 1987.
[Huan91] Huang, S. c., & Huang, Y. F. "Bounds on the number of hidden neurons in
multilayer perceptrons,", In IEEE Transactions on Neural Networks Vol 2,
pp 47-55, 1991.
[Hush93 J Don R. Hush and Bill G. Horne, "Progess in Supervised Neural Networks:
What's New Since Lippmann?", In IEEE SIGNAL ProceSSing magazine, pp
8-39, January 1993.
[Ishi90] M. Ishikawa, "A structural learning algorithm with forgetting of link
weights," Tech. Rep. TR-90-7, Electroteclmical Lab., Tsukuba-City, Japan,
1990.
[Jac088] Adaptation." Neural Networks, Vol 1, pp 295-307.
[Ji90] C. Ji, R.R. Snapp, and D. Psaltis, "Gereralizing smoothness constraints
from discrete samples," Neural Commputation, Vol. 2, no. 2, pp. 188-197,
1990.
[Kam90] E . D. Karnin,"A simple procedure for pruning back-propagation trained
neural network," IEEE Trans. Neural Networks, Vol. 2, no. 2, pp. 239-242,
1990.
[Krus88] Kruschke, J, "Creating local and distributed Bottlenecks in Hidden Layers
of Back pop gat ion Networks," in Touretzky, D., Hiton, G., and Sejnowski,
T (Eds): Proceeding oj the Connectionist Models Summer School 1988,
Morgan Kaufman Pub!., San Mateo, California, 1988.
[Makh89] 1. Makhoul, A. EI-Jaroudi, and R. Schwartz, " Fonnation of disconnected
decision regions with a single hidden layer." In Proceedings oj the
International Joint Conference on Neural Networks. Vol 1, pp 455-460,
1989.
[Merz88] Merzenich, M.M., Recanzone, G., Jenkins, W.M., Allard, TT, and Nudo,
R)., "Cortical Representional Plasticity," In Rakie, P. and Singer, W.
(Eds) : Neurobiology of Neocortex. John Wiley and Sons Limited. S.
Bernhard. Dahlem Konferenzen, 1988.
[McIn89] McInerny, J.M., K.G. Hainer, S. Biafore, and R. Hecht-Nielsen (1989).
"Back Propagation Error Surfaces Can Have Local Minima.", In
International Joint Conference on Neural Networks (Washington 1989),
vol. II, 627. New York:IEEE.
[Mins69] Minsky, M.L. and S.A. Papert, "Perceptron," Cambridge: MIT Press.
Partially reprinted in Anderson and Rosenberg, 1988.
[Morg91J D.P. Morgan and c.L. Scofield, Neural Networks and Speech Processing.
Kluwer Academic Publishers, 1991.
[Moze89] M. C. Mozer and P. Smolensky, "Skeletonization: A technique for trimming
the fat from a network via relevance assessment,", In Advances in Neural
Information Processing (1), D.S. Touretzky, Ed. (Denver 1988), 1989, pp.
107-115.
[Mura91] K.Murase, YMatsunaga and Y.Nakade: "A Back-propagation Algorithm
with Automatically Determines the Number of Association Units", Proc. of
International Joint Conference on Neural Networks(IJCNN-SINGAPORE),
I, p.783-788, 1991.
[pe1i93J M. Pelillo and A. M. Fanelli, "A method of pruning layered feed-forward
neural networks," in Proc IWANN'93, Sitges, Barcelona, June 1993
(Berlin: Springer-Verlag).
[Reed93] R. Reed, "Pruning A1goritluns-A Survey", in IEEE Transactions On
Neural Network, Vo14, NO.5, September 1993, pp. 740-747.
[Rume86] Rumelhart, D.E., James McClelland, "Parallel Distributed Processing, ", Vol
1 and 2, MIT Press, Cambridge, MA, 1986.
[Siet88] Sietsma., 1. and R.lF. Dow (1988), "Neural Net Pruning-Why and How."
In IEEE International Conference on Neural Networks (San Diego 1988),
Vol. 1,325-333. New York:IEEE.
[Tsai93] Tsaih, R., "The Softening Learning Procedure." In Mathematical and
Computer Modeling, Vol. 18, No.8, 61-64.
[Tsai93] Tsaih, R., "The Softening Learning Procedure for The Layered Feedforward
Networks With Multiple Output nodes". Proceeding of
Intemational Joint Conference on Neural Networks, Nagoya, I, 593-596.
[Wata93] Watanabe E. and Shimizu H., "Algorithm for Pruning Hidden Nodes in
Multi Layerd Neural Network for binary StimulusClassification Problem",
In Proceedings of 1993 Intemational Joint Conference on Neural
Networks, pp 327-330, 1993 .
[Weig91] A. S. Weigend, D. E. Rumelhart, and B. A. Huberman, "Generalization by
weight-elimination with application to forcasting," in Advances in Neural
Information Processing (3), R. Lippmann, J. Moody, and D. Touretzky,
Eds, 1991, pp. 875-882.
[Werb90] P. J. Werbos, "Backpropagation Through Time: What It Does and How to
Do It", In Proceedings of the IEEE, Vol 78, No.1 0, October 1990, pp .
1550-1560.
[Werb74] P. Werbos. IIBeyond Regression :New Tools for Prediction and the Analysis
in the Behavioral Sciences." Ph.D. thesis, Harvard, Cambridge, MA,
August 1974.