Share Email Print

Proceedings Paper

Dynamic neural network for fuzzy inference
Author(s): Madan M. Gupta; George K. Knopf
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

An innovative design of a dynamic neural network architecture that is able to first learn and then utilize fuzzy-like `IF-THEN' rules is presented in this paper. Each fuzzy neuron in the network represents a compositional rule of inference that defines the relationship between a particular premise and the corresponding consequence. The neural network first determines the similarity between a neural input (a discretely sampled fuzzy set) and the feedforward synaptic weights (accumulated knowledge-base). The `best' definition of the input is selected by competition arising from the dense feedback between the neurons. A satisfactory conclusion is reconstructed in weighted feedforward outputs from the `winning' neuron. The knowledge-base is updated by an unsupervised learning algorithm that adapts the feedforward weights assigned to both the neural inputs and outputs. An example of how this dynamic neural network can be used to perform fuzzy-like inference rules for the navigation of an autonomous vehicle through an unstructured environment is used to illustrate these notions.

Paper Details

Date Published: 22 December 1993
PDF: 14 pages
Proc. SPIE 2061, Applications of Fuzzy Logic Technology, (22 December 1993); doi: 10.1117/12.165050
Show Author Affiliations
Madan M. Gupta, Univ. of Saskatchewan (Canada)
George K. Knopf, Univ. of Saskatchewan (Canada)

Published in SPIE Proceedings Vol. 2061:
Applications of Fuzzy Logic Technology
Bruno Bosacchi; James C. Bezdek, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?