Also @tdietterich that’s a strawman; I am not calling for hand-coded knowledgebases; I am calling for hybrid systems that include symbol knowledge, some core parts innate, but also plenty of learning.
yes, i do believe human brains have innate capacity for manipulating operations over variables, see eg my 1999 infant experiments reported in science, extended to newborns by Gervain, plus theoretical arguments in Algebraic Mind
-
-
but i also believe that symbol manipulation can be implemented i neural networks; whole point of algebraic mind was to consider solutions
-
The majority of DL community will tell you that this 'algebraic mind' can be trained via gradient descent. In fact, that's exactly what the NLP community is doing. Graph Networks have explicit structural features that aid reasoning. Time leads to richer methods.
- 13 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.