And the whole idea that is being advocated is that the degree search can be implemented overtop a discreet search space (...Boolean inference can be *relaxed* into probabilistic inference). That relaxation has to mean the probabilistic inferences are going to be referring to overlapping inferences doesn't it? Or at least something that can -in most cases- be seen as roughly equivalent to overlapping references as far as the potential for complexity is concerned.
Jim Bromer On Wed, Apr 23, 2014 at 2:51 PM, Jim Bromer <[email protected]> wrote: > I don't think that the probabilistic inference is (necessarily) linear if > the interval probabilities may overlap or interact with other intervals in > a non-sequential way. I think that even non-sequential methods are not > linear if the method of search-backtracking is needed. > > Jim Bromer > > > On Wed, Apr 23, 2014 at 12:05 AM, YKY (Yan King Yin, 甄景贤) <[email protected] > > wrote: > >> On Sat, Apr 12, 2014 at 4:49 PM, Anastasios Tsiolakidis >> <[email protected]>wrote: >> >>> Disclaimer: the answer is probably several hours or days of googling >>> away, just trying to save time here. >>> >>> I haven't looked into (non probabilistic) inference engines for a long >>> time, and it looks like the landscape has changed. I consider them >>> indispensable for game-like domains (including physics), and I would like >>> to hear about the performance and subtleties of the new breed. I am working >>> towards general game playing set-ups, which I've mentioned before as a kind >>> of AGI drosophila. >>> >> >> >> May not answer your question, but boolean inference can be relaxed into >> probabilistic inference with interval probabilities, and that becomes a >> linear programming problem (with linear constraints on probabilities). The >> latter can be solved in polynomial time. >> >> Probabilistic inference is actually more desirable than boolean >> inference, as far as AGI is concerned. So the fixation on the NP hardness >> of boolean inference may be unnecessary. On the other hand, probabilistic >> inference may offer a route to tackle the P=?NP question. >> >> As for boolean inference, I am not aware that the landscape has changed >> drastically (as far as new techniques are concerned, but some new software >> may have emerged). The key issue here is to choose the kind of logic, for >> example whether you want propositional logic, description logic (as used by >> Semantic Web technologies), first order logic, or higher order logic. The >> state-of-the-art engines for each kind of logic are constantly improving, >> but seems to be slowly =) >> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >> <https://www.listbox.com/member/archive/rss/303/24379807-f5817f28> | >> Modify<https://www.listbox.com/member/?&;>Your Subscription >> <http://www.listbox.com> >> > > ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
