From: Josh Gardner
Subject: Heteroassociative backprop networks in LISP..?
Date: 
Message-ID: <36F73CF8.DA182EBD@dowco.com>
Greetings all..this is a cross posting from comp.ai.neural networks I
hope you don't mind. I figured that, since I am looking for some LISP
oriented help, I might as well post here too.


Greetings all.

I'm having trouble understanding how to create, and, more to the point,
train a heteroassociative backprop net. I understand auto-associative
nets, and
I can even build a hetero-associative one, but I can't train it.

Perhaps some background would help, so you see my level of knowledge.
The code I have been playing with is sample code, in LISP, given to me
by one of my profs. It only trains in an autoassociative way. Basically,
I could build say, a 4-2-4 backprop net, and it will train properly, on
whatever data I decide to feed it.

Now, let's say I decide to use the program to train a 4-2-2 backprop
net. it iwll still try to train it in the same manner and a 4-2-4, and
obviously, I
don't get teh output I'd like.

So, my question to all you wise folk out there is, could any of you
suggest where I might find some LISP source code that can train networks
in a
heteroassociative, that I might better understand the differences
between the two in code, and that I might get a better understanding of
how to code
one myself? Unfortunately, Neural Networks are not a large portion of my
course, and so we never covered this material in class. Thus, I am
seeking out other sources.

Thanks very much! Replies are welcome by email, or posted to this group.

Josh