neural resonator..any help?

more here..
http://www.jayemsonic.de/resoneuroEngl.html

This is something I have been interested in for quite some
time. Using neural nets in Max that is...

A neural network is a network of weighting (learning)
functions that have a persistent weighted state. The weighting
function determines what impulses will pass through. When you
start out, your neural net is a blank slate. You have to feed
it data in order to train it.

This article suggests you can train the neural net with audio
feedback which is a very interesting concept. What is not clear
is what the learning function would be used. If the function just
pass the value through, you are not really "learning" anything.
I would be interested to know more about how different learning
functions could be applied and how that would change the flow of
data in the diagram (pic 9). His suggestion of using genetic
algorithms is interesting, but not really "learning".

I think there are all sorts of interesting things you could
explore with neural nets. As far as I know there are no Max
externals out there that implement a neural net. Please let me
know if there is because that would save me a lot of time.
If anybody knows of a good general purpose neural net library,
I would definitely port it.

I think CNMAT made an external, but I cant seem to find it right now. I remember having it and not knowing what the heck to do with it. Ive seen others as well, at least one other.

http://www.maxobjects.com/?PHPSESSID=d7068802cc3c611d8ee0ec937c1f3147&PHPSESSID=d7068802cc3c611d8ee0ec937c1f3147&request=neural&operateur=AND&id_plateforme=0&Submit=OK

heres my search on maxobjects.com looks like there's several. Id love to see a patch made with one that I could study.

Anthony Palomba wrote:
> His suggestion of using genetic
> algorithms is interesting, but not really "learning".

GAs have been used with neural nets as ways of 'evolving' optimal
network topologies for various problems, which, AFAIK, correlates with
what we know about learning in the cognitive, embodied sense (that
brains aren't fixed networks of nodes).

I didn't read the blurb fully yet, but didn't get much of a neural net
flavour (in the sense of learning structures) so much as a Gordon
Mumma's cybernetic console kind of vibe. All very interesting ...

--
Owen

hi Anthony,
i know a bunch of max objects-patches that deal with something similar but i've never been able getting something (to be honest i did not even understand some of these!)..

http://www.olliebown.com/main_blog/?page_id=8

http://www.neuromuse.org/downloads/index.html

best
mic

Quote: Anthony Palomba wrote on Thu, 08 May 2008 18:40
----------------------------------------------------
> This is something I have been interested in for quite some
> time. Using neural nets in Max that is...
>
> A neural network is a network of weighting (learning)
> functions that have a persistent weighted state. The weighting
> function determines what impulses will pass through. When you
> start out, your neural net is a blank slate. You have to feed
> it data in order to train it.
>
> This article suggests you can train the neural net with audio
> feedback which is a very interesting concept. What is not clear
> is what the learning function would be used. If the function just
> pass the value through, you are not really "learning" anything.
> I would be interested to know more about how different learning
> functions could be applied and how that would change the flow of
> data in the diagram (pic 9). His suggestion of using genetic
> algorithms is interesting, but not really "learning".
>
> I think there are all sorts of interesting things you could
> explore with neural nets. As far as I know there are no Max
> externals out there that implement a neural net. Please let me
> know if there is because that would save me a lot of time.
> If anybody knows of a good general purpose neural net library,
> I would definitely port it.
>
>
----------------------------------------------------

hmmmm, very interesting. I will check it out.

Quote: Axiom-Crux wrote on Thu, 08 May 2008 09:52
----------------------------------------------------
> I think CNMAT made an external, but I cant seem to find it right now.

Here's our funny webpage about that:

http://archive.cnmat.berkeley.edu/MAX/neural-net.html

This project is dead.

mz

Too bad there is only a mac version...

I have an mxj Java class at:

http://miajo.co.uk/?MaxMSP:mxj_nnLists

... though I haven't had time to check it with Max 5 (I know there are some mxj issues as someone reported my mxj curve~ not working).

Works in Max 4.6.3 though, and I'll get round to it soon.

For info: This is about 4x slower than a C version by my calculations (the Java code was ported from some of my neural net C code using a similar algorithm).