mouss wrote:
>Shelby Moore wrote:
>> 
>> Sorry but that lack of technical detail on our web site about the patent 
>is intentional for now.
>> 
>
>then why not wait for the patent to be pending/filed before starting a
>discussion?


Because the purpose discussion in the other thread which you are referring to 
was not concerned with the patentability of AccuTechnology:

http://marc.theaimsgroup.com/?l=spamassassin-devel&m=110979008703837&w=2


The point of that discussion was to determine how to integrate AccuTechnology 
into SA at this time (from now to summer).  It was to get the harmonious 
direction figured out.

>You have the full right to believe that you found _the_ solution. 
>Unfortunately for you, you aren't the first one. did you read the FUSSP 
>doc at
>       http://www.rhyolite.com/anti-spam/you-might-be.html


I know.  He wrote that because of me.   Google on "kook".

>you might also wanna check the litterature about "text categorization", 
>"information retrieval", ...


We have.


>You might be a genius,


Thanks but I am not.


> but the probability of someone being a genius is 
>ridiculously low (except from his viewpoint and that of his 
>family/friends/... but that's called auto-whitelistring:-). so please 
>give us the right to doubt.


Agreed.  I expect lots of doubt.  I still doubt.  Everyday I strive to prove 
more.  As you said, it takes time to get from point a to point b.


>tell you what? I am writing a paper that will provide a mathematical 
>proof that no method can get 0% FP and 0% FN,


Agreed.

Caveat: AccuSpam's (not AccuTechnology) Daily Summary can achieve 0% FPR, 
because the missed spam is sent as summary only.


> and more, that any method 
>can only get "better" at some expenses. (no, I can't give you any infos 
>unless it is finished:-)


Agreed.  In our case some of the "expenses" are that we have to collect MUCH 
MORE data and we can not train on non-bulk spam, so we still need Bayesian for 
the non-bulk spam.


>Anyway, if you believe in your solution, implement it as a proxy so that 
>all projects can use it. and good luck.


Thanks.


>as to the nfw reports, please keep'em for marketers and investors. 
>everybody knows what benchmarketing is. An inferior system that is tuned 
>for a test situation will always outperform a superior system used out 
>of the box.


Tuning cost is part of the cost equation.

However, I agree that nfw report was probably biased towards the Postini 
provider they use normally.

However, I think it is known fact that a system like BrightMail can achieve 
about 95% without any training and very, very few FPs.   I am not sure if SA is 
significantly better without training.  I think AccuTechnology is an 
"order-of-magnitude" better, but I don't have enough data to prove it.  I can 
only say with confidence that it is much better than other bulk correlators 
such as Razor and DCC.



>
>to shorten the matters:
>- I don't believe your solution is good (not because I hate your 
>firstname or your domain, but because I have no idea what it really is).


You mean you doubt.  Understood.


>- I personally don't wanna see anything commercial directly implemented 
>inside SA, unless it can be used according to SA license (we already 
>have enough licenses to read). I understand that you might not make a 
>life out of that, but fortunately for me, that would be your life. I 
>only care of mine (I'm kidding of course, I'll love to help but only if 
>I feel I can).


No problem.  We can emulate the Razor license, if that is already acceptable to 
SA.

>anyway, good luck.

Thanks!  Appreciated.  After all, the desire is to kill spam.

You know what got me started on anti-spam????

I was running DownloadFAST.com and our customers often couldn't buy because 
they could get their password via email (we verify email as anti-fraud check 
and to insure contact before capturing the charge on the card account).

I got so frustrated with false positives, that I decided to contact Vernon at 
DCC (major mistake!) to see if we could improve to a system that was more 
accurate.  Well the rest is history.  Just Google.

Regards,
Shelby

Reply via email to