The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9564

Title:Analog Neural Networks with Nonmonotonic Transfer Functions
Date: 1995
Abstract:The recollection properties of certain autocorrelation associative memories are studied in this thesis. The models under consideration include the classical Hopfield network, as well as the networks with nonmonotonic transfer functions described by Morita and Yoshijawa. The results are mainly of an experimental nature, and are compared with the corresponding theoretical predictions. In most cases, the two have been found to agree, but there are some remarkable exceptions. Along with this study, the classical results in the area of autocorrrelation associative memory are summarized and reviewed, with an emphasis in the dynamical propertis of the recurrent network model, Liapunov functions and the Cohen-Grossberg theorem.

[Search These Pages] [DAI Home Page] [Comment]