Abstract
A neural network is developed to mimic a human brain. The neural network consists of units and links that connect between units. Various types of neural networks are categorized into two classes: (1) back-propagation hierarchical neural network and (2) mutual-connected neural network. Generally speaking, it is hard to fix the number of units to build a neural network for solving problems. So the number of units is decided on the basis of experts’ experience.
In this paper, we explain a learning method how to decide the structure of a neural network for problems. The learning method is named structural learning. Even if we give a sufficient number of units, the optimal structure will be decided in the process of learning.
The objective of the paper is to explain the structural learning of both hierarchical and mutual connecting neural networks. Both networks obtained and showed the sufficiently good results. In the stock forecast by a general neural network, the operation and the system cost are very large because a lot of numbers of hidden layer units in the network are used. This research tried the optimization of the network by the structured learning, and evaluated the practicality. ...
Keywords
- Neural Network
- Hide Layer
- Stock Price
- Stock Prex
- Portfolio Selection
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
This is a preview of subscription content, access via your institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsPreview
Unable to display preview. Download preview PDF.
References
Arisawa, M., Watada, J.: Enhanced Back-Propagation Learning and Its Application to Business Evaluation Proceeding. In: IEEE Interational Conference on Neural Network, Orlando, Florida, pp. 1155–1160 (1994)
Asahi, T., Murakami, K., Sagamihara, T.: BP Algorithm for Optimality and highly speeding of Neural Network Structural Learning. Report of Institute of Electronics, Information, NC90-64 (1991)
Elman, J.L.: Finding Structure in Time. Tech.Report CRL-8801, University of California, San Diego (1988)
Hayashi, Y.: Stock Price Is Forecastable - Introduction to Chart Analysis (in Japanese). Kanki Publishing, Tokyo (2000)
Jordan, M.I.: Serial order: A parallel, distributed processing approach. Tech.Report, ICS-8604, University of California, San Diego (1986)
Matsunaga, Y., Nakade, Y., Nakagawa, O., Murase, M.: Buck Propagation Algorithm to Auto-Elimitation of hidden units in Higher Neural Network. Trans. of Institute of Electronics, Information, J74-D-II(8), 1118–1121 (1991)
Matsunaga, Y., Nakade, Y., Nakagawa, O., Kawanaka, M.: Buck Propagation Algorithm to auto eliminate redundant hidden units from Competition. Trans. of Institute of Electronics, Information, J79-D-II(3), 403–412 (1996)
Murata, I., Watada, J., Choi, J.: Structural Learning of Neural Network and Its Application to Stock Forecasting. In: CD Proceedings, International Conference on Management Engineering (ISME2006), Kitakyushu, Japan, March 10-12, R30001-R30006 (2006)
Watada, J., Oda, K.: Formulation of a Two-layered Boltzmann Machine for Portfolio Selection. International Journal of Fuzzy Systems 2(1), 39–44 (1999)
Watada, J.: Keynote Speech on Structural Learning of Neural Network and Its Application. In: Proceedings, An International Conference on Intelligent systems (ICIS2005), Kuala Lumpur, Malaysia, Dec. 1-3 (2005)
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 2007 Springer Berlin Heidelberg
About this paper
Cite this paper
Watada, J. (2007). New Perspective for Structural Learning Method of Neural Networks. In: Melin, P., Castillo, O., Aguilar, L.T., Kacprzyk, J., Pedrycz, W. (eds) Foundations of Fuzzy Logic and Soft Computing. IFSA 2007. Lecture Notes in Computer Science(), vol 4529. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-72950-1_24
Download citation
DOI: https://doi.org/10.1007/978-3-540-72950-1_24
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-72917-4
Online ISBN: 978-3-540-72950-1
eBook Packages: Computer ScienceComputer Science (R0)