Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uwm.edu!csd4.csd.uwm.edu!markh
From: markh@csd4.csd.uwm.edu (Mark William Hopkins)
Newsgroups: comp.ai.neural-nets
Subject: Re: Connectionist Finite State Machines -- description of an architecture
Message-ID: <7984@uwm.edu>
Date: 1 Dec 90 00:30:56 GMT
References: <7982@uwm.edu>
Sender: news@uwm.edu
Organization: University of Wisconsin - Milwaukee
Lines: 11

In article <7982@uwm.edu> markh@csd4.csd.uwm.edu (Mark William Hopkins) writes:
>   (d) The same process is applied to the hidden input layer and input layer to
>       derive the activations of the output layer.
>
>An alternative to (d), would be to simply pass the hidden layer's activations
>to the output layer, but this will not be guaranteed to produce valid learning
>behavior though it is a simpler architecture.

A minor correction: the hidden layer should STILL be connected to the output
layer in either case, or there won't be any way for error signals to propagate
back to train the state transition function.