[pdp-discuss] normalizing factor?

[Randall C. O'Reilly]

By default this value is the target kwta.pct value on the layer: i.e., the k/n 
or pct value set in the kwta setting in the layerspec.  In the most recent 
versions of leabra (avail as version 3.2aX from grey ftp server), there is a 
flag on the kwta params to set a different actual layer activity level, if 
for some reason it differs from the target value.  In older versions you 
could set this in the conspec.  You can definitely compensate by setting 
*absolute* (not relative) scaling: this normalization enters as an absolute 
factor..

- Randy

On Monday 08 May 2006 17:13, Brian Monroe wrote:
> Hi again,
>
> Another question, this time regarding the normalization factor based on
> "expected activity level of the sending projection" (alpha_k, in equations
> 2.15 and 2.16 in "the book") in leabra.  I doesn't appear that the
> calculation of this factor is made explicit anywhere.  How precisely is
> this calculated?
>
> Also, it doesn't appear there is an option to set this as a parameter
> anywhere.  Am I correct, or is alpha_k one and the same as "savg" that can
> be fixed?  And if it is not tunable, I would like the option to compensate
> for alpha_k with the absolute and/or relative projection scaling parameters
> to get limited or no normalization across projections.  Will this work?
>
> Thanks,
>
> 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