Learning potential function and differential inclusion

Author

Xiong, Momiao ; Wang, Ping

Author_Institution

Georgia Univ., Athens, GA, USA

Volume

1

fYear

1992

fDate

7-11 Jun 1992

Firstpage

401

Abstract

A unified mathematical theory of neural learning is presented. A learning potential function for a neural network is introduced, and a novel dynamical system approach to nondifferentiable, global optimization problems is proposed. A differential inclusion (DI) for finding a global minimum of a learning potential function is derived. Asymptotic results for the solutions to these DIs are obtained. A consistency theorem for parameter estimation is proven. Applications to supervised learning and unsupervised learning are investigated

Keywords

neural nets; optimisation; parameter estimation; unsupervised learning; consistency theorem; differential inclusion; dynamical system approach; global optimization problems; neural learning; neural network; parameter estimation; potential function learning; supervised learning; unified mathematical theory; unsupervised learning; Artificial neural networks; Differential equations; Information management; Information processing; Information technology; Neurons; Random access memory; Statistics; Supervised learning; Technology management;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 1992. IJCNN., International Joint Conference on

Conference_Location

Baltimore, MD

Print_ISBN

0-7803-0559-0

Type

conf

DOI

10.1109/IJCNN.1992.287178

Filename

287178