123 points by quantum_researcher 6 months ago flag hide 16 comments
johnsmith 6 months ago next
This is quite interesting! I've been following the developments in neural networks compression and this seems like a major breakthrough. Looking forward to seeing how it performs in real-world applications.
proggie 6 months ago prev next
I've read the paper and I'm blown away! This could significantly reduce the computational cost of deploying NNs in resources-constrained environment. Great job!
johnsmith 6 months ago next
I completely agree, it's a game changer.
bigdata 6 months ago prev next
Has anyone tried implementing it yet? Looks too good to be true.
machinelearner 6 months ago next
I have, it works as advertised. The code is open-sourced on Github, I suggest you to give it a try.
deeplearning 6 months ago prev next
The idea of using lossy compression techniques in neural networks is not new but the authors have certainly taken it to the next level. Good job!
neuralnetworks 6 months ago next
Couldn't agree more, the results are quite impressive.
ganalyst 6 months ago prev next
This is certainly promising but I wonder how it performs when applied to GANs and other complex architectures.
proggie 6 months ago next
From what I've read, it's still an open research question.
mathgeek 6 months ago prev next
The math behind this paper is quite involved, I had to read it multiple times to understand it.
newbie 6 months ago prev next
I'm new to this field but I'm trying to learn. Any resources that you'd recommend for someone who wants to get started with NNs compression?
machinelearner 6 months ago next
Checkout the book 'Deep Learning' by Goodfellow, Bengio, and Courville. It has a section dedicated to the topic.
bigdata 6 months ago prev next
Any thoughts on how it could be applied to dimensionality reduction?
neuralnetworks 6 months ago next
It could potentially be used to perform matrix factorization and extract low-dimensional structure from data.
csstudent 6 months ago prev next
I'm working on a similar project for my computer science course, any advice on how to not get overwhelmed by the math and get results faster?
mathgeek 6 months ago next
Start with implementing a simpler model, gradually move towards more complex ones. Also, use plenty of visualizations to understand the data better.