It’s basically data parallelism. Moore’s law is diminishing.
The computing cost of deep learning scales with the ninth power of performance improvement. We can’t go on like this.https://spectrum.ieee.org/amp/deep-learning-computational-cost-2655082754 …
-
-
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
I agree that pursuing performance increases using bigger and bigger models does not make much sense. However, we have many compression techniques (quantization, pruning, etc) that significantly reduce deep learning resource requirements when using middle size models.
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
/q A clever breakthrough might find a way to make deep learning more efficient or computer hardware more powerful, which would allow us to continue to use these extraordinarily flexible models. So the race is on for clever breakthroughs, which will probably be forthcoming.
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
yes, investments in deep learning research need to go towards either training data reduction and/or towards facilitation of domain adaptation.
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
as said before, we won't make a nonlinear time jump in science until we understand natural intelligence ...
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
Thank you, Professor Great Reading
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
What are some other methods that provide similar or better performance with cheaper computational cost than DL... Some great points mentioned by Yann on the topic.pic.twitter.com/RmK0mcRmxX
-
We don’t have them yet, which is why we need more research. BTW, this piece by Yann is a very weak defense of DL - he knows it’s hideously inefficient, and doesn’t know what to do about it.
Keskustelun loppu
Uusi keskustelu -
-
-
We can make fusion energy or reduce the power consumption of GPUs. Or rethink the whole neural network approach. My question is why deep learning work? The explainability of deep models is unsatisfactory at the moment.
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
this is a fascinating thought. i think it shows some modesty in the way we can automate intelligence. personally i am convinced by task oriented AI for long there is no question this is already in action
-
but then it means we do not know how biological systems compute with so little energy compared to what AI systems need
- Näytä vastaukset
Uusi keskustelu -
Lataaminen näyttää kestävän hetken.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.