simple NN in Perl (was RE: LISP program)

[Brian Densmore]

Well ,I don't know perl, but let me give a stab at my
analysis of your neural net. It looks like you have 
built a single layer neural network. These aren't 
terribly useful. A three layer neural net would be
a minimum for really useful and interesting things
(although there are certain uses for two layer ones).
Three layers seem to be a bare minimum for decent
learning. Secondly you didn't include any weighting
to individual nodes nor any way to alter the weighting
on a per node basis. Nor is there any way for your neural net
to learn. But it is a basic shell you've built
that could be expanded on and if it were an object could
be a base to build multiple layers on. Also you've only
built two state nodes. You could add a fuzzy aspect to
it to make a node partially on. Or you could add a 
disease to the nodes to make some neurons die/behave abnormally.
But I digress, yes, you've got the beginnings of a neural net,
since all they are all matrices of numbers in the end. But it
is lacking in a great many respects. Yes any paradigm can be done in 
any language, some are better suited than others, but I still
stick to my statement that hardware neurons are better. Wonder if
you can still buy them?

> -----Original Message-----
> From: david nicol 
> Sent: Monday, February 23, 2004 11:29 AM
> To: Brian Densmore
> Cc: kclug
> Subject: simple NN in Perl (was RE: LISP program)
> 
> 
> On Fri, 2004-02-20 at 14:23, Brian Densmore wrote:
> 
> > Neural networking is best done with hardware.
> > But I used C++ to build neural networks in college.
> 
> 
> Any paradigm can be implemented in any language, provided
> that the implementor has sufficient understanding of both.
> Because I think it willbe easy and I want to show off I am
> going to present a Perl framework for a neural network system,
> later in this message, simulating simultaneity by providing
> time slices to all nodes completely at random, and deferring
> implementation of the connections and activity of each node. Nah,
> I'll provide a stub that attempts to average numbers.
> 
> Here goes:
> see attachment.  pipe it through less to see the nodes
> reach a consensus, sort of.
> 
> 	perl perlNN.pl | less
> 
> Brian, does this match the sort of thing you were doing in C++?
> 
> 
> 
> -- 
> david nicol
>                                                   Ralph Nader 
> eats cats
>