questions of inmates and symbol manipulation are only fully separable if one can point to a system that can actually learn the basics of symbol manipulation (eg variables and operations over variables, per The Algebraic Mind). AFAIK nobody has ever done thishttps://twitter.com/neurograce/status/1003341028609482752 …
-
The neural program synthesis community has been working on neural + symbolic methods for programming (and by extension, symbol manipulation) for some time now. There has been some progress; see https://arxiv.org/abs/1611.01989 , https://arxiv.org/abs/1703.05698 , https://arxiv.org/abs/1611.01855 .
i cites some other papers of same sort in my january pieces, critical appraisal and defense of skepticism
-
-
Your essays are inspiring. That said, those of us advocating hybrid models also need to identify/publicize concrete tasks where these models clearly beat DL. You've mentioned Flashfill-like inductive programming. Another one: learning functions with formal guarantees. What else?
-
alphaGo is a hybrid. also my guess is that the best driving systems are hybrids, but i don’t know if any good comparison w end to end deep learning, with so much being proprietary.
- 1 more reply
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.