Implementing static links in SNNS

• Messages sorted by: [ date ][ thread ][ subject ][ author ]
• Next message: : "Re: Test functions for Jordan Elman Nets."
• Previous message: : "recording activation of hidden units"

I am currently attempting to implement a Model Based Neural Network
(Ref: Caelli, Squire and Wild (1993), Neural Networks 6(3)). 

The basic idea is to reduce the complexity of the weight search process
by encoding a-priori knowledge into the network. This is achieved by
instantiating some of the weights in the neural network with specific
values that are not affected by the training process. Other weights in
the network can be extracted (once training is completed) as values
meaningful to the initially encoded heuristic.

Has anyone already attempted to implement such a network in SNNS? If
not, does anyone know how (or if) this can be achieved in SNNS? 

Many thanks
Russell Keith-Magee

-- 
Russell Keith-Magee BSc(Curtin)
Honours student and Graduate Research Assistant
School of Computing, Curtin University of Technology
Snail Mail: GPO Box U1987, Perth WA 6845
email: keithmag@cs.curtin.edu.au
Phone (Work): (08) 9266 3481

• Next message: : "Re: Test functions for Jordan Elman Nets."
• Previous message: : "recording activation of hidden units"