Re: 2 questions

["Randall C. O'Reilly" <oreilly@grey.colorado.edu>]

"Luke Jerzykiewicz" <ljerzyki@ccs.carleton.ca> writes:
> (1) I'm trying to replicate a network, published in a recent article, =
> which uses  "a Gaussian activation function that ranges between 0 and 1, =
> which a SD of 1.  The [hidden] units generate a maximum response of 1 =
> when their net input is equal to the mean of the Gaussian."  I can't =
> substitute the standard sigmoid function in my replication because I'm =
> interested in analyzing properties of the trained hidden layer and that =
> would defeat the point.  Can what I'm trying to do be done in PDP++?  =
> How do I substitute the gaussian for the sigmoid function?  Looking at =
> page 21 of the manual makes me pessimistic.

Support for Gaussian (RBF) functions is in the so++ code, and could be
adapted for other architectures.  See docs on so in the manual.

> (2) I've tried to find the various example networks for use with =
> Computational Explorations in Cog-Neuroscience but without much luck.  =
> I"ve used both Windows and PDP++ to try to locate them.  I'm not =
> entirely sure that the standard downloadable package includes the =
> examples at all.  Am I wrong about that?  Where are they?  Do I need to =
> download them separately?  What's the URL?

They are a separate download from the basic pdp++ code:

http://psych.colorado.edu/~oreilly/cecn_download.html

				- Randy