Looks like Geoff finally got his capsules to work 
https://nips.cc/Conferences/2017/Schedule?showEvent=9167 …
-
State-of-the-art performance on MNIST!
I don't understand why one would invest so much time in developing an algorithm, then demo it on MNIST
-
-
When a paper reports results on MNIST it usually means that they tried it on other datasets, and it didn't work. It's still fine to publish.
-
I don't think it's fine to only list experiments on toy datasets, and hide failures on more complicated cases.
- Show replies
New conversation -
-
-
I would be expecting reviewers to state that eval on cifar, mnist etc. alone indicates problems with scalability & result quality. Funny ...
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
sometimes new and underdeveloped ideas need time to mature and scale
-
such undercooked ideas should then be workshop papers not conf. While it is subjective what qualifies, common toy problems aren't enough
- Show 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.