RE: Neuronet applications?

[djl <djl@xxxxxxxxxxx>]

I've done a few neural network analysis examples, and have one or two 
programs that would apply (brianmaker, NeuroGenetic Optimizer)   What's 
done is to take the icon or other pixelated character representation and 
train the net to recognize it. This means forming pixel images of many 
examples of each character.  Then the formed neural net is used to test an 
example of the unknown. It's resemblance to the known character is thus 
quantified, and "belonging" can be established.   I think, first, that 
there are so few variants that a person highly familiar with the text could 
"beat" the neural net.  The fun might be in assessing A and B variants.
Don

On Sunday, April 09, 2000 3:40 PM, John Grove [SMTP:jgrove@xxxxxxxxxxx] 
wrote:
>
> 	I'm venturing into a field that I know extremely little about here... A
> neighbour of mine mentioned that there are plenty of Neuronet
> back-propogation character analysis applications on the market that
> should be able to take a large amount of 'odd characters' and determine
> with a fair amount of confidence - what is or what is not a single
> character.
>
> 	Does anyone know anything about neuronets and back-propogation, and
> whether it has any potential use in the analysis that we're trying to
> do?
>
> 	John.
>
> 4groves@xxxxxxxxx
> john@xxxxxxxxxxxx
> jgrove@xxxxxxxxxxx