I am using the ANN external. It works pretty well, but it only implements Kohonen Feature Map type ann's. I like those because you can get output while training. So you can start from scratch and let the performance untill time T be the learning set for the decision at time T.
What I am looking for is a way to use variable dimension feature vectors. I would like to run different lengths of rhytmic patterns into a single neural net. hmm, now that I am thinking about it, I could use some kind of Goedel numbering scheme to map multidimensonal data to 1 dimension. Or I could use different projections on hyperplanes. Back to the computer to try this out, thanks for the question :)
BTW if the above makes no sense, please ask, there are some resources on the net for neural nets, but I found it confusing when I first started looking into this.
Gerard
From: "matthew s shanley" arkadyan@lycos.com
Does anyone have any experience using neural network computing with PD or anything similar? I want to experiment with an external or something using neural networks for decision making in response to incoming sounds played by me. I'm willing to explore, but since I have no experience with writing externals, nor with neural networks, any help or starting points would be appreciated.
matt
Hi Gerard,
I like those because you can get output while training.
is it not the only way to train it in a neural network process ?
Any info about it is very useful to me, you could forward us the link on the internet. I personally don' t know the ANN external so if you could give me the link as well... Thanks. On Wednesday, June 18, 2003, at 03:47 pm, vanDongen-Gilcher wrote:
I am using the ANN external. It works pretty well, but it only implements Kohonen Feature Map type ann's. I like those because you can get output while training. So you can start from scratch and let the performance untill time T be the learning set for the decision at time T.
What I am looking for is a way to use variable dimension feature vectors. I would like to run different lengths of rhytmic patterns into a single neural net. hmm, now that I am thinking about it, I could use some kind of Goedel numbering scheme to map multidimensonal data to 1 dimension. Or I could use different projections on hyperplanes. Back to the computer to try this out, thanks for the question :)
BTW if the above makes no sense, please ask, there are some resources on the net for neural nets, but I found it confusing when I first started looking into this.
Gerard
From: "matthew s shanley" arkadyan@lycos.com
Does anyone have any experience using neural network computing with PD or anything similar? I want to experiment with an external or something using neural networks for decision making in response to incoming sounds played by me. I'm willing to explore, but since I have no experience with writing externals, nor with neural networks, any help or starting points would be appreciated.
matt
PD-list mailing list PD-list@iem.at http://iem.at/cgi-bin/mailman/listinfo/pd-list
Alex said at "Re: [PD] Re: neural networks and pd."r[2003/06/18 13:48]
Hi Gerard,
I like those because you can get output while training.
is it not the only way to train it in a neural network process ?
Yes and no.Kohonen maps are self adjusting. With other types of neural nets you first give a training set of inputs and desired outputs to train the network.
Gerard
-
Alex -
vanDongen-Gilcher