[pdp-discuss] question about unit-group based connectivity

[Brian Monroe]

Apologies for continuing the conversation on the list, but this seems like something someone else might find useful at some point...

So you could do a work-around by using different layers instead of unit groups (to get the different con specs), but can you get cross-layer kwta inhibition to act identically to within-layer inhibition?



Brian Monroe
Graduate Student
University of Southern California


----- Original Message -----
From: Randall O'Reilly <oreilly at psych.colorado.edu>
Date: Friday, March 31, 2006 10:01 pm
Subject: Re: [pdp-discuss] question about unit-group based connectivity
To: Brian Monroe <monroe at usc.edu>
Cc: pdp-discuss at psych-srv3.Colorado.EDU

> Brian,
> 
> These Prjn specs are for creating different patterns of 
> connectivity  
> between units according to unit groups, not for different patterns 
> of  
> weight values -- that is actually not that easy to do without 
> writing  
> some new C++ code (a new ConSpec) because there is one conspec per  
> projection, and so you'd have to create a different projection for  
> each unit group to achieve this using the standard stuff -- doable, 
> 
> but not very efficient.  Cheers,
> 
> - Randy
> 
> On Mar 31, 2006, at 1:50 PM, Brian Monroe wrote:
> 
> >
> > I was having a hard time figuring out how to implement a layer 
> with  
> > unit groups with their own random weight initializations/ 
> > distributions.  Am I correct in assuming I need to be using a  
> > GpFullPrjnSpec (or possibly a GpOneToManyPrjnSpec)?  If this is 
> the  
> > case, I did not see any of the example projects that use these  
> > specs, nor did I see it discussed in the book, and so it is not  
> > clear how to actually implement them.  If this is not the case,  
> > what is the easiest way to set up these sub-distributions of  
> > weights within a layer?
> >
> > Thanks for any info.
> >
> > Brian Monroe
> > Graduate Student
> > University of Southern California
> > _______________________________________________
> > PDP-Discuss mailing list
> > PDP-Discuss at psych-srv3.Colorado.EDU
> > http://psych-srv3.Colorado.EDU/mailman/listinfo/pdp-discuss
> 
>