YKY said: "Multi-layer perceptron is a very inspiring case because it made use of the sigmoid function to allow differentiation to solve the learning problem via gradient descent. The latter is a very efficient algorithm. So I'm looking in that direction." --------------------------------------------- But not all learning can take place via gradients. There are too many things wrong with the presumption to believe that some kind of network learning modality was all that is missing to make it work. I am not just being negative.
Jim Bromer On Wed, Jan 28, 2015 at 4:44 PM, YKY (Yan King Yin, 甄景贤) via AGI < [email protected]> wrote: > On Wed, Jan 28, 2015 at 1:10 AM, Matt Mahoney via AGI <[email protected]> > wrote: > >> YKY, is this building on your last 10 years of work on Genifer? What >> problems did you encounter that require a fundamental redesign? >> > > In logic-based AI, the central algorithm is proof-search which is > combinatorial search. This cannot be avoided as long as formulas are > symbolic and discrete. Although heuristics can be designed to speed up the > search, such techniques are not very inspiring in the sense that they don't > use advanced maths. (Though there is still a possibility that classical AI > heuristics are sufficient to bootstrap AGI.) > > Multi-layer perceptron is a very inspiring case because it made use of the > sigmoid function to allow differentiation to solve the learning problem via > gradient descent. The latter is a very efficient algorithm. So I'm > looking in that direction. Deep learning is fashionable now and it can > potentially lead to techniques capable of learning complex structures. > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/24379807-653794b5> | > 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
