Re: Renormalization and Contrast Enhancement in CPCA

["Neil Thompson" <n.thompson@roehampton.ac.uk>]

Thank you for this Randy. This does partially solve the problem -I'd missed
that errata- , though I'm still not able to reproduce the Leabra behaviour
as I increase wt_off beyond about 6 (for wt_gain =6, savg_cor=1) or
decreasing savg_cor below 0.7 (for wt_gain=6, wt_off=1) - I was able to
improve the matching however by fixing the value of the sending unit average
activation, rather than computing it through the simulation.

Neil

----- Original Message -----
From: "Randall C. O'Reilly" <oreilly@grey.colorado.edu>
To: <n.thompson@roehampton.ac.uk>
Sent: Sunday, February 01, 2004 3:58 PM
Subject: Re: Renormalization and Contrast Enhancement in CPCA


> Neil,
>
> Have you corrected your equation based on the errata posted on the
> book website?  I suspect this is the problem:
>
> 7/30/03: Ch 4, p. 135, equation 4.23. Lee Newman @ Univ Michigan noted
> that the Theta (wt_off) parameter should be in the denominator of the
> w/(1+w) ratio, not the numerator as it is written. Thus, it should
> read: (w_ij / (\Theta (1 - w_ij))^-\gamma instead of: (\Theta w_ij /
> (1 - w_ij))^-\gamma.
>
> from: http://psych.colorado.edu/~oreilly/cecn_errata.html
>
> Hope this helps,
>
> - Randy
>
> -------------------------------------------------------
> This message has been scanned for viruses and dangerous
> content by The University of Surrey Roehampton
> MailScanner, and is believed to be clean.
>
>


-------------------------------------------------------
This message has been scanned for viruses and dangerous
content by The University of Surrey Roehampton
MailScanner, and is believed to be clean.