Skip to Main Content
Frequently Asked Questions
Submit an ETD
Global Search Box
Need Help?
Keyword Search
Participating Institutions
Advanced Search
School Logo
Files
File List
case1055427027.pdf (3.22 MB)
ETD Abstract Container
Abstract Header
On the trainability, stability, representability, and realizability of artificial neural networks
Author Info
Wang, Jun
Permalink:
http://rave.ohiolink.edu/etdc/view?acc_num=case1055427027
Abstract Details
Year and Degree
1991, Doctor of Philosophy, Case Western Reserve University, Electrical Engineering.
Abstract
This thesis presents a systems-theoretic approach to analysis and synthesis of artificial neural networks for system modeling and optimization. The thesis consists of two independent parts under a unified framework. The roles of artificial neural networks in both descriptive and prescriptive aspects are investigated via formalization, categorization, and characterization. Starting with defining concepts and attributes, this thesis formalizes a general framework of artificial neural networks. Based on the formalization, this thesis characterizes the artificial neural networks by analyzing their asymptotic properties such as trainability, stability, representability, and realizability. Specifically, the conditions of potential trainability, absolute trainability, and global and absolute trainability are derived; statistic implication of supervised learning for representability of ANNs are discussed. Conditions of asymptotic stability, feasibility and optimality of solutions to optimization problems are also derived. Based on the characterization, the thesis derives the design methodology for synthesizing architectures and rules of artificial neural networks. For feedforward ANNs, the issues of size and the desirable topological structure of network architectures are a ddressed; the criteria to select training set and evaluation function are addressed, a dynamical configuration rule, a synchronous-asynchronous learning rule, and a training paradigm are proposed. For recurrent ANNs, guidelines for determining evaluation rule, activation rule, aggregation rule, and penalizing procedure are also discussed.
Committee
M.D. Mesarovic (Advisor)
Pages
115 p.
Keywords
artificial neural networks
Recommended Citations
Refworks
EndNote
RIS
Mendeley
Citations
Wang, J. (1991).
On the trainability, stability, representability, and realizability of artificial neural networks
[Doctoral dissertation, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1055427027
APA Style (7th edition)
Wang, Jun.
On the trainability, stability, representability, and realizability of artificial neural networks.
1991. Case Western Reserve University, Doctoral dissertation.
OhioLINK Electronic Theses and Dissertations Center
, http://rave.ohiolink.edu/etdc/view?acc_num=case1055427027.
MLA Style (8th edition)
Wang, Jun. "On the trainability, stability, representability, and realizability of artificial neural networks." Doctoral dissertation, Case Western Reserve University, 1991. http://rave.ohiolink.edu/etdc/view?acc_num=case1055427027
Chicago Manual of Style (17th edition)
Abstract Footer
Document number:
case1055427027
Download Count:
544
Copyright Info
© 1991, all rights reserved.
This open access ETD is published by Case Western Reserve University School of Graduate Studies and OhioLINK.