As I understand it, FOL is only Turing complete when predicates/relations/functions beyond the ones in the data are allowed. Would PLN naturally invent predicates, or would it need to be told to specifically? Is this what "concept creation" does? More concretely: if I gave PLN a series of data, and asked it to guess what the next item in the series would be, what sort of process would it employ?
Thanks, --Abram Demski On 8/4/08, Ben Goertzel <[EMAIL PROTECTED]> wrote: > On Mon, Aug 4, 2008 at 6:10 PM, YKY (Yan King Yin) < > [EMAIL PROTECTED]> wrote: > >> On 8/5/08, Ben Goertzel <[EMAIL PROTECTED]> wrote: >> > >> > 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 ... >> >> >> Yes, I think the inductive search is somewhat similar to backward >> chaining, >> except that the steps in the inductive search can *create* rules, whereas >> in >> backward chaining you're applying *existing* rules. >> > > > Yes, but in PLN/ OpenCogPrime backward chaining *can* create hypothetical > logical relationships and then seek to estimate their truth values > > See this page > > http://opencog.org/wiki/OpenCogPrime:IntegrativeInference > > and the five pages linked to from it (at the top) > > >> >> We need a scoring function, but I have not thought about this yet. >> >> I think the hardest part is actually in generating the search tree. You >> see, in first-order logic, rules can involve many predicates, predicates >> may >> have variables as arguments, and the arguments may even have complex terms >> involving functions. So the combinatorial explosion is severe. >> > > The purpose of the scoring function is precisely to attempt to manage this > combinatorial explosion. > > >> >> The scoring function may provide a "gradient" over the search space, so >> you >> suggested to use hill-climbing. But I suspect that such a gradient is not >> useful during the search, because the search space is discrete and >> irregular, and the scores probably jump irregularly from node to node. >> That's why I suspect that hill-climbing is not useful here. >> > > As noted in one of the pages mentioned above, > > http://opencog.org/wiki/OpenCogPrime:HebbianInferenceControl > > I believe that the only solution to this problem is not algorithmic, but > architectural: we need to mine the data-store of historical inferences > > http://opencog.org/wiki/OpenCogPrime:InferencePatternMining > > and use this information to provide inductive bias to be used within the > scoring function itself. > > -- Ben G > > > > ------------------------------------------- > 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/?&; > Powered by Listbox: http://www.listbox.com > ------------------------------------------- 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
