Gaussian blur around target states

[Tim Rogers <tim.rogers@mrc-cbu.cam.ac.uk>]

Hi folks,

	I've got a simple recurrent networks whose input and output layers represent 
the position of an object in space on an x-y plane.  To make the units 
representing adjacent points in space similar to one another, I'd like the 
input activation and target states to bleed over to neighbouring units; that 
is, to represent a point at X,Y (for inputs and targets), I'd like to have the 
unit at X,Y on the layer grid maximally active; and nearby units 
(topographically speaking) somewhat active.  Is there some quick-and-dirty way 
to do this in pdp++?

	Thanks for any help,

	TR

--
Dr. Timothy Rogers
MRC Cognition and Brain Sciences Unit
15 Chaucer Road, Cambridge, UK
CB2 2EF

tel:+44 1223 355 294 xt. 360
fax:+44 1223 359 062
email: tim.rogers@mrc-cbu.cam.ac.uk