Discussions related to Visual Prolog
augustinharrison
Active Member
Posts: 29
Joined: 28 Jul 2010 1:28

learning algorithm in pie

Unread post by augustinharrison »

Hi Thomas Linder Puls,
In the early 90's my brother and I took the pie and created an abstraction of an expert and built the Enterprise Knowledge Server or EKS. With this inference, interpreter, processor our company was able to build knowledge into sql tables and run complex business rules as an interpreted set of knowledge definitions. We sold an iteration of this to HBOC which became McKesson who then dominated the medical billing industry and remains number 1 to this day. This has made a lot of money and has VIP running Contract Medical Billing in 500 sites. We are now talking to Dell about building with AI principles using the EKS. I need to emphasize learning capabilities that are built into the pie as well as what can be implemented. My thinking has been that opening a file, adding rules or facts and saving that file is a form of learning but that is too simplistic. I would like to promote that we have learning built into our AI.
Can you cite any examples of learning algorithms in pie? A demo with Dell already resulted in a leading magazine ranking us in the top 20 AI software companies.

Thanks, A.E. Harrison
aeh
User avatar
Thomas Linder Puls
VIP Member
Posts: 1318
Joined: 28 Feb 2000 0:01

Re: learning algorithm in pie

Unread post by Thomas Linder Puls »

It sounds like you know more about this than me.

To my knowledge the current "boost" in machine learning is solely based on neural network technologies.

I think of neural networks as "statistic intelligence", and "expert systems" and the like as "analytical intelligence".

When it comes to "analytical learning", where you try to deduce "rules" from scenarios, I believe that there has been a lot of university work in this field. But it is not very clear to me if any of this has ever lead to something practically useful. I think a kind of basic problem is that you can often infer contradicting rules and how do you deal with that.

Neural networks (~statistics) could provide an "intuition" for dealing with such contradictions (and for other things). And I am certain that car companies, google, facebook, etc combine statistical and analytical approaches in order to obtain better results than the approaches can do individually. But it seems that everybody is holding their cards close to their chest.

We have had some people looking a little at neural network technologies, but we have not created bindings to any such things. And we are certainly not at a level to advise in combining the approaches.
Regards Thomas Linder Puls
PDC
augustinharrison
Active Member
Posts: 29
Joined: 28 Jul 2010 1:28

Re: learning algorithm in pie

Unread post by augustinharrison »

Thank you for that response. You are the highest authority with vip prolog in my view. i Am hoping perhaps a rule with weights saved after being invoked in a process can be mildly considered a learned analytical result.

AEH
aeh
Victor Yukhtenko
Posts: 9
Joined: 3 May 2001 23:01

Re: learning algorithm in pie

Unread post by Victor Yukhtenko »

Hi AugustinHarisson,

If you wish, you may contact me (victor@pdc.spb.su).
I am either interested in this area.
We are working on the Vip-based platform, where PIEngine is one of the plugins. It may communicate with other Vip-based plugins.

regards,
PDC-SPb
Victor Yukhtenko.
Post Reply