HELP BPEDITMENU                             Julian Clinton  Feb 1990


This file describes the options available on the Edit Back-propagation
Network Menu.


         CONTENTS - (Use <ENTER> g to access required sections)

 -- 'learning rate (eta)'
 -- 'learning momentum (alpha)'

-- 'learning rate (eta)' ----------------------------------------------

Changing the learning rate allows a back-propagation network to learn at
a greater or lesser rate. A large value of eta allows a network to learn
more quickly but may cause  the network to oscillate without  converging
on a solution. A small value of  eta will make this less likely but  can
increase the learning time manyfold.


-- 'learning momentum (alpha)' ----------------------------------------

The learning momentum (referred to as alpha) defines how the change in
strength of a particular connection is affected by the previous change.
This can prevent a network becoming stuck around a local minima. However
if the momentum is too great, the network might begin to oscillate
around the global minima.


--- Copyright Integral Solutions Ltd. 1990. All rights reserved. ---