The problem of not being able to recreate and adequately compare the performance of many publications is real. However, a "reproducibility crisis" similar to psychology in a field of mathematical engineering would mean fraud, and I'm not convinced yet that much fraud is going on.https://twitter.com/SilverJacket/status/964219816944037889 …
-
-
Replying to @ferrouswheel
I agree that authors need always to be prepared to fully demonstrate how they got their result. But if ML was like psychology, then it would turn out that GANs don't work, ImageNet challenges are mostly done with selected random guesses etc.
1 reply 0 retweets 1 like
Replying to @Plinz @ferrouswheel
At the moment, it might be that some proudly reported 2% advantages over competing algorithms turn out to be flukes, but it is unlikely that they just don't work, or a whole family of models is going to collapse under scrutiny, as it happens in psychology.
8:58 PM - 16 Feb 2018
from Cambridge, MA
0 replies
0 retweets
1 like
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.