I think that learning from a single experience is feasible when you either have a sophisticated innate reaction to generates a kind of innate recognition or you already have a good insight into some of the factors (or components) that occurred during the situation. In some cases the insights may be morphed from similar learning that might not fit perfectly but can be shaped to fit some aspects of the singular occurrence. Jim Bromer
On Mon, Dec 10, 2012 at 6:11 PM, Aaron Hosford <[email protected]> wrote: > I have some things to add: > > - One-shot generalization. Even animals can try something once and > remember never to do it again. Every learning algorithm I've ever seen > failed at this. Most of them require tens of thousands of iterations to get > the point. This is because learning algorithms are looking for convergent > estimates of expected reward, and converging adjustments in reward have > limited utility in the real world. > - Ability to learn from communication. Human intelligence isn't just > the intelligence of the individual. It's the collective intelligence of the > species. I don't have to learn things the hard way. I can just ask someone > if something's a good idea, or what sort of approach they recommend. Why > reinvent the wheel at every step? > - Going back to my recent replies to Jim: Effective caching of > relevant information. Knowledge is cached information, whether gleaned from > observation or computation. Once again, why reinvent the wheel at every > step? > > > > On Sat, Dec 8, 2012 at 3:40 PM, Piaget Modeler > <[email protected]>wrote: > >> Once we have a good set of requirements we can begin a design by finding >> design >> elements that match the requirements, assumptions, dependencies, and >> constraints. >> >> *Requirements*: >> >> 1. Efficient Organization >> 2. Efficient Execution >> 3. Easy to Understand >> 4. Supports vast interconnections among concepts >> 5. Rapid Execution to continually reevaluate multiple paths >> 6. Need to switch search spaces rapidly >> 7. Need to expand or restrict search spaces dynamically >> 8. Need for concept integration >> 9. Need for concept differentiation >> 10. Need to fluidly recombine concepts >> 11. Need to support simulation and multiple path exploration. >> 12. Supports justification discovery / Reason-based reasoning and >> planning. >> 13. Should be able to explain reasons for decisions. >> >> >> >> *Assumptions* >> >> >> *Dependencies* >> >> >> *Constraints* >> >> >> >> Anything else to add? Are any of these Assumptions or constraints rather >> than requirements? >> Kindly advise. >> ~PM. >> > ------------------------------------------- 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
