that describes an unfortunate number of my models :|
Next you can try the Deception layer: works like a regular layer at training time, but its output may be random at test time --sometimes.https://twitter.com/fhuszar/status/786194531943186437 …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
or a MisconceptionLayer where random dropout is replaced by averaging weights at test time pretending E[f(w)]=f(E[w]) Oh, wait...
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fhuszar ImmaculateConceptionLayerThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fhuszar can it be used to boost ensemble performance ?Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I think the discriminator network in a GAN should be called antideception net and the generator deception net
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
or maybe works like a contraception at training time, but a identity layer at test time. And use it for NSFW models!
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.