Very really revealing thread, as DL field try immunizes itself from candid talk about it limits -- by saying both that (A) we already know the limits (@ylecun) but also that (B) that there are no known limits (@roydanroy, @earnmyturns).
These can't both be true.https://twitter.com/GaryMarcus/status/1173377561897816065 …
-
-
It's a long list, since it's akin to describing the frontier of research. A few: 1) meta-learning requires too many tasks (and where do they come from?) 2) fine-tuning for transfer learning has no guarantees 3) RL requires a reward function, which is unrealistic for many tasks
-
yes, indeed, making such a list does indeed define the frontier of research, which was precisely the point of http://Rebooting.AI . Since our list doesn't overlap with yours, I will be curious to see which of those on our list you do and do not agree with. [+1 to yours.]
- 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.