New perspective for structural learning method of neural networks

Junzo Watada

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    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.

    Original languageEnglish
    Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Pages231-240
    Number of pages10
    Volume4529 LNAI
    Publication statusPublished - 2007
    Event12th International Fuzzy Systems Association World Congress, IFSA 2007 - Cancun
    Duration: 2007 Jun 182007 Jun 21

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume4529 LNAI
    ISSN (Print)03029743
    ISSN (Electronic)16113349

    Other

    Other12th International Fuzzy Systems Association World Congress, IFSA 2007
    CityCancun
    Period07/6/1807/6/21

    ASJC Scopus subject areas

    • Computer Science(all)
    • Biochemistry, Genetics and Molecular Biology(all)
    • Theoretical Computer Science

    Fingerprint Dive into the research topics of 'New perspective for structural learning method of neural networks'. Together they form a unique fingerprint.

    Cite this