Mailing List Archive

SA/MAD Interfacing
Considering that my Linode can't deep-learn anything or become any more
intelligent than it already is, it seems reasonable that I'd have to
interact with an external system.

Does anyone know of any RFCs or other standards that define how
Machine-learning, Artificial intelligence, and Deep-learning systems
communicate with Email hosts?



Thanks,

-- Jared Hall
Re: SA/MAD Interfacing [ In reply to ]
On 10/1/2021 3:27 AM, Marc wrote:
> That is indeed a good futuristic vision. That small mail providers cooperate as a group to exchange information and facilitate a distributed compute environment to analyze and block spam.
> I can remember asking here 1-2 years ago someone about shareing a bayes db or so (can't remember), and his reaction was totally the opposite.
>
1)  Thanks, but not that original.  Extrapolated from figures 4 and 5 in
https://www.sciencedirect.com/science/article/pii/S2405844018353404
2)  I don't think ISP/ESPs cooperate that much.  They might participate
if there are controls over the metadata shared with a non-competitive
third-party.
3)  I've done sharing of Bayes successfully.  I certainly would not do
it with auto-learning and manual retraining has it's obvious problems. 
Last I looked at the Bayes modules, the token expiration times were
pretty fast.

My thinking is more aligned with MAD developers/researchers versus
ISP/ESPs.  Much of what passes as MAD "research projects" are simply
regurgitated versions of Bayes.  Everybody tests with the same corpora
however old it may (or may not) be.  Real-world use data is practically
non-existent.

Since SA rule-based systems have such great efficiency, perhaps a better
use-case of ML/AI/DL would be in Virus/Malware control. Still, I
envision an environment where the ISP/ESP can set privacy flags, what
services are requested (AV/Spam), and what algorithm (or algorithms) to use.

-- Jared Hall








>
>> -----Original Message-----
>> From: Jared Hall <jared@jaredsec.com>
>> Sent: Friday, 1 October 2021 06:51
>> To: users@spamassassin.apache.org
>> Subject: SA/MAD Interfacing
>>
>> Considering that my Linode can't deep-learn anything or become any more
>> intelligent than it already is, it seems reasonable that I'd have to
>> interact with an external system.
>>
>> Does anyone know of any RFCs or other standards that define how Machine-
>> learning, Artificial intelligence, and Deep-learning systems communicate
>> with Email hosts?
>>
>>