Re: Spam

[Matt Curtin <cmcurtin () interhack net>]

"Mosher, Robert E." <robert.mosher () cmet af mil> writes:

> How hard would it be to train a neural net to recognize spam?

Not terribly difficult.  Enough spam has been collected that
performing the analysis needed to build a few layers of recognizers
shouldn't be much of a big deal.  The real question is whether a
neural net would be inherently better than a score-based system.

I'd be inclined to believe that although a scoring system would
probably be more expensive computationally (because in practice, all
tests will probably be performed, as opposed to a neural net, where
it'll take a path and perform a small number of tests), it'd probably
also be more effective.

Of course, you could build examples of efficient or inefficient
implementations either way.

-- 
Matt Curtin  Interhack Corp  +1 614 545 HACK http://web.interhack.com/
Author,  Developing Trust: Online Privacy and Security  (Apress, 2001)
Knight, Lambda Calculus | Certum quod factum. --Giovanni Battista Vico
_______________________________________________
firewall-wizards mailing list
firewall-wizards () nfr com
http://list.nfr.com/mailman/listinfo/firewall-wizards