Monday, February 9, 2009

New Algorithm From Dr. Forgy

Greetings:

Yes, that's right, folks.  Dr. Forgy is working on a new algorithm that, as Diana (his wife who is also a computer programmer from CMU) puts it, "Blows the old Rete 2 / Rete III algorithm right out of the water."  No, it is not ready yet.  There are still several parts to work out and it may not be completed until the end of the year.  

I talked with Dr. Forgy yesterday afternoon about that (and some other things) and he gave me permission to disclose it.  He said that the new algorithm (as yet unnamed) should be about 10 times faster (hopefully) than the Rete 2 / Rete III Algorithm.  And, the really cool part is that he will be discussing this at the October Rules Fest.   What he will discuss and how much will be revealed is still to be determined since it won't be finished (again, hopefully) until the end of summer, just before the October Rules Fest.

The name?  He doesn't have one yet.  I suggested CLiF Algorithm (his initials are CLF) but he thought that was too much ego and, if you have ever met the man, he has a small ego and is a totally "nice guy" - something unusual in our day and time.  But, having given away the original Rete Algorithm (it was his Ph.D. thesis done under a DoD grant) I think that he would like to retain the license on this one and have some kind of royalty for the use of it.  As you may or may not know, Rete 2 / Rete III is licensed ONLY to Fair Isaac under his agreements when FIC bought Rules Power way back when.

Bottom line:  (1) New Algorithm from Dr. Forgy.  (2) Dr. Forgy will be discussing this at October Rules Fest this year.  (3) The new algorithm "blows Rete 2 out of the water".  :-)  (4)  The new ORF 2009 site will be up and running this week so that you can register early and take advantage of the "Early Bird" discount.

SDG
jco

2 comments:

woolfel said...

Wow, sounds cool. If there was anyone that could think of a way to speed up pattern matching, it would be Dr. Forgy. Lately, I've been working on a topology cost function for pattern matching networks. The idea itself is quite simple. I hope to have a draft done in the next 3-4 weeks. I've already blogged about it and provided basic description.

Bob Schlicher said...

Yet another good reason to go to the October Rule Fest.

Glad to hear about this and looking forward to hearing more.

BobS