Basically Gary Marcus articulated the same unease I have with broad applications of ML: https://arxiv.org/ftp/arxiv/papers/1801/1801.00631.pdf …
-
-
Show this threadThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
For production CG, ML denoising trained on the particular scene assets in use can work well. A hand-wavey defense is that it provides priors for better estimating the integrals from the samples you can afford. It biases the result but hopefully in a helpful direction
-
Yeah…I know it works, but I wish we could solve the problems more holistically…
End of conversation
New conversation -
-
-
it's a pretty unfortunate hack on NLP also sometimes fwiwhttps://twitter.com/bblum0/status/997857082840928256 …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
if by soft problems you mean approximation problems, then yes
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I feel the same way when I read about applying ML to operating systems and especially file systems! (ML for scheduling could be interesting though...)
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
This Tweet is unavailable.
-
-
It seems reasonable to use ML to find an algorithm for an exact solution, no? I agree it’s unsatisfying when ML approximations are used though.
-
(ML might not be good enough at these sorts of things yet though. I’m unfamiliar with the problem domain.)
End of conversation
New conversation -
-
-
What are your thoughts on the Learned Indices paper? While the idea isn’t fleshed out adequately, it seems to contradict your unease, I think https://arxiv.org/abs/1712.01208
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.