An interesting thought that comes to mind is that Neural Painters only required a single GPU colab notebook to run! If you had access to a huge amount of compute, do you think you would have produced better research work? It's something I ask myself quite often...
-
-
-
I like thinking that my constraints force me to try unusual things, but it's unclear to me if unusual == better.

- Još 1 odgovor
Novi razgovor -
-
-
I am really not trying to defend the overuse of compute infrastructure (I agree that we don't go into the right direction), but on the other hand we wouldn't make this kind of remarks to guys from CERN on reproducing the Higgs Boson experiments either, right?
-
My point of view: the research on efficient networks and model compression seems to be too disconnected from mainstream research on deep learning. We find compressed models and low compute training mostly only in papers in which efficient compute is the actual topic of research.
- Još 3 druga odgovora
Novi razgovor -
-
-
Run their code on 1 gpu and multiply the results times 16.Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
16 GPUs is really not so bad, but that they can simply buy Richard Sutton if they want to try a little reinforcement learning is outrageous...
- Još 2 druga odgovora
Novi razgovor -
-
-
Depends on the situation. If it’s model-free RL then it sounds about right. I never did that stuff before joining DeepMind and it was quite frustrating to learn that almost nothing can be trained on my workstation (I did my best and failed miserably). It is what it is, I guess
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I wish people were upfront about the fact that their code is a waste of time for computer setups that cost categorically less than used cars, when hardware assumptions are so thoroughly baked into the code that it will run on nothing less than originally programmed for.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
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.
