I mentioned earlier that I'd forward a private email I'd previously sent to YKY, on the topic of probabilistic inductive logic programming.
Here is is. As noted there, my impression is that PILP could be implemented within OpenCog's PLN backward chainer (currently being ported to OpenCog by Joel Pitt, from the Novamente internal codebase) via writing a special scoring function ... -- Ben G ---------- Forwarded message ---------- From: Ben Goertzel <[EMAIL PROTECTED]> Date: Fri, Jun 6, 2008 at 9:27 AM Subject: Re: logical implication (was: modus ponens) To: "YKY (Yan King Yin)" <[EMAIL PROTECTED]> Hi, >> By the way what do you think of this work >> >> http://people.csail.mit.edu/kersting/ecmlpkdd05_pilp/pilp.pdf >> >> This is "probabilistic inductive logic programming" and seems like it >> would fit well into OpenCog at first glance, though I didn't read it >> over carefully yet. > > Very good paper, I'm very interested in this stuff and will study it > carefully. However I can only give you some brief comments at this > time. The paper talks about several ways of defining probabilistic > logic learning, and I will decide which one makes the most sense. > > The algorithms in the paper will work for small KBs, but I'm almost > 100% sure that they won't work for large KBs. So, they may be of > experimental value in OpenCog. It seems like one of their algorithms involves a scoring function, score(), which basically plays a quite similar role to the heuristic fitness function used in the PLN backward chainer. I'm not sure how different what they're doing is from PLN backward chaining, actually.... The whole art, then, is in estimating the score/fitness function associated with each possible inference step... > My most ambitious goal is to figure out how to do "probabilistic + > fuzzy ILP for very large KBs". IMO the essence is in the scoring/fitness function that guides pruning, which must be set via statistical experience based on prior inferences... ... >snikt< ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=108809214-a0d121 Powered by Listbox: http://www.listbox.com
