Maybe every paper abstract should have a mandatory field of what the limitations of the proposed approach are. That way some of the science miscommunications and hypes could maybe be avoided.
Seen that @techreview piece on "neural net solves the three-body problem 100 million times faster”?
here are some counterpoints from my collaborator Ernie Davis:
re https://www.technologyreview.com/s/614597/a-neural-net-solves-the-three-body-problem-100-million-times-faster/ …
#toyproblemspic.twitter.com/3lJnnp5DKK
-
-
- 1 more reply
New conversation -
-
-
There’s nothing intrinsically wrong with toy problems. On the contrary they can be exceptionally useful in research. The problem is solving hard problems require exceptional domain knowledge to be able to think about what toy problems might be useful. 1/2
-
I don’t know nearly enough about the three-body problem to comment on this. However, it so happens I know more than most people in the world about a number of big challenges being tackled by DL researchers. Sad to say, a severe lack of domain knowledge abounds. 2/2
End of conversation
New conversation -
-
-
I suppose the next step is to train nets that can generalise from the 3-body to higher body problems. That'd be extremely useful.
- 2 more replies
New conversation -
-
-
Tangent: This is why I like Ex Machina so much, because the conditions of the test were built around a very specific solution (decided ahead of time). That kind of pass or solution is only good for bragging rights.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.