spf-discuss
[Top] [All Lists]

Re: A hyperdimensional anti-spam machine?

2004-08-25 15:37:24

Great summary!  I surmise you would also need to map the multi-dimension 
surface of non-spam and make sure you were below the surface of that to not 
risk a false positive.  In other words, since you are correlating variates that 
have cross-correlation between spam and non-spam, then there are two surfaces 
(spam and non-spam) which can intersect.  If your email is caught between the 
two surfaces, then you have NEUTRAL answer.

The Laplacians I believe are used to approximate the surface in useful way for 
this measurement.


On Wednesday 25 August 2004 09:07 am, Holm, Mark wrote:
I know this is OT, but anybody even tangentially involved in spam
fighting ought to read this article.
http://zdnet.com.com/2100-1107_2-5323437.html?tag=zdfd.newsfeed

I won't mention it again on this list.


Unfortunately, I will mention it. ;-)

This machine isn't hyperdimensional - the vectors it calculates are. (Turn 
on your multi-variable geometry brain now.) The idea is you have multiple 
characteristics of a piece of mail - whether it passes SPF, whether it 
passes a Bayesian filter, etc, etc... If you create a vector for each mail, 
where each coordinate of that vector is the score it gets based on that 
measurement, then you can determine whether the email is spam or not based 
on whether the vector is above the surface of spam or below the surface of 
spam. The trick lies in being able to determine if a piece of mail is 
definitively above or below the surface of spam based on only one or two 
characteristics. Hyperdimensional means "more than 3 dimensions". It sounds 
wild and crazy, but in something as esoteric as determining if a bunch of 
bytes is spam or not, it is appropriate.

- -- 
Jonathan M. Gardner
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)

iD8DBQFBLQ/jBFeYcclU5Q0RAvp6AKCGalxBYhPAn2Blpp/T9F3Vl2DfxQCeN99a
kic4V1I2qWpBI4ENrTS9QzQ=
=8FLW
-----END PGP SIGNATURE-----

-------
Sender Policy Framework: http://spf.pobox.com/
Archives at http://archives.listbox.com/spf-discuss/current/
http://www.InboxEvent.com/?s=d --- Inbox Event Nov 17-19 in Atlanta features 
SPF and Sender ID.
To unsubscribe, change your address, or temporarily deactivate your 
subscription, 
please go to 
http://v2.listbox.com/member/?listname=spf-discuss(_at_)v2(_dot_)listbox(_dot_)com


<Prev in Thread] Current Thread [Next in Thread>