How can I select the centers of the first hidden layer of nn with the orthogonal least squares algorithm????

Hi, I'm designing a radial basis neural network. I want to select the centers of the first hidden layer with orthogonal least square. I have seen qr decomposition in matlab, but How can I use it for my problem????HELP!!!

Answer 1

[Greg Heath]

MATLAB's NEWRB neither

1. Allows a specified starting configuration

2. Allows RBFs with different widths.

For either, you have to modify the function. The source code is available via

type NEWRB

You may be better off looking at earlier versions of the code. They contain less boilerplate overhead.

There is a lot of pertinent info on NEWRB that is accessible by searching the NEWSGROUP and comp.ai.neural-nets using

NEWRB

If you decide to modify it keep in touch.

Good Luck.

Thank you for accepting my answer

Greg

[Greg Heath]

http://www.mathworks.com/matlabcentral/newsreader/view_thread/151286#888395

is a 30 post thread based on how to best use NEWRB. Tedious. But if you are going to use NEWRB, definitely worth reading.