[FRIAM] FW: National Science Foundation Update Daily Digest Bulletin

[Marcus G. Daniels]

Nicholas Thompson wrote:
> I am too dumb to know the degree to which I am being kidded here.  
> Please explain..
Suppose 100 people give 999 responses to yes/no questions and all of 
them answer by flipping a coin.  A final answer correctly answers the 
question "Are your eyes blue?"  Just by chance, amongst those 999 coin 
flips some can be weakly correlated to the eye color question and linear 
combinations of them may turn out to be even more correlated (as there 
are more bits for encoding, bogus covariation though it is).  So 
sometimes there is a need to generalize or `regularize' high dimensional 
data to reduce overfitting.   A simulation is potentially one way to do 
regularization.  Another example is using an `average face' for face 
recognition:

http://www.sciencemag.org/cgi/content/abstract/319/5862/435
>  
>  
>
>     ----- Original Message -----
>     *From:* Douglas Roberts <mailto:doug at parrot-farm.net>
>     *To: *nickthompson at earthlink.net
>     <mailto:nickthompson at earthlink.net>;The Friday Morning Applied
>     Complexity Coffee Group <mailto:friam at redfish.com>
>     *Sent:* 2/15/2008 8:41:18 AM
>     *Subject:* Re: [FRIAM] FW: National Science Foundation Update
>     Daily Digest Bulletin
>
>     Run that lousy data through a simulation, and then publish the
>     results as truth.
>
>     Works every time!
>
>     --Doug
>