A network that uses the outer product rule, hidden neurons, and peaks in the energy landscape

Author

Hoffmann, Geoffrey W. ; Davenport, Michael R.

Author_Institution

Dept. of Phys., British Columbia Univ., Vancouver, BC, Canada

fYear

1990

fDate

1-3 May 1990

Firstpage

196

Abstract

The results of the development and simulation of an extended version of the discrete Hopfield neural network are summarized. The network uses hidden neurons to optimize the orthogonality of the memory space. The process is fast because it is noniterative, and the design is such that a hardware implementation would require no executive processor during memory storage or retrieval. Simulations indicate that the storage capacity of the network and the radius of attraction of each memory are significantly better for uncorrelated memories than those of the standard Hopfield model. Hidden neurons permit flexibility in the network capacity for memories of a given length and make it possible for the network to solve second-order problems

Keywords

content-addressable storage; memory architecture; neural nets; discrete Hopfield neural network; energy landscape; extended version; hidden neurons; memory space optimisation; outer product rule; second order problem solving; simulation; Analog circuits; Councils; Hardware; Hopfield neural networks; Intelligent networks; Least squares approximation; Neural networks; Neurons; Physics; State-space methods;

fLanguage

English

Publisher

ieee

Conference_Titel

Circuits and Systems, 1990., IEEE International Symposium on

Conference_Location

New Orleans, LA

Type

conf

DOI

10.1109/ISCAS.1990.111966

Filename

111966