By unnoticed you mean published for just a year?https://arxiv.org/abs/1812.11118
-
-
- Još 3 druga odgovora
Novi razgovor -
-
-
Interesting indeed. This is something I’ve noticed with many deep models, and glad to know it wasn’t my setup :) (although I still cannot be sure)
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
The double descent seems to only occur within the region of non-zero training error. If you consider that most of deep learning attempts to be in the zero-error region, it may still not be that relevant for deep learning.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I have seen versions of this quite often in time series
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I wonder if the double descent is a rediscovery of the peaking phenomenon?http://37steps.com/2448/trunks-example/ …
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
How deep is the model size rabbit
hole? Why not bigger models than 60+... where is the limit?Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Interesting. I've noticed this in multiple applications and always attributed that to some an initial naive solution of the network. For example, in semantic segmentation when objects are small, a easy naive guess could be to predict everything as background. etc.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Not really! There were many works in the past, especially from Opper, Advani and more recently from Belkin et. al. demonstrating this behavior!
- Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.