Exploring the Depths of Neural Network Pruning with Lottery Ticketing(medium.com)
65 points by triketora ago | | | 13 comments
65 points by triketora ago | | | 13 comments
hacker1 4 minutes ago | prev | next
This is a really interesting exploration of neural network pruning! I wonder how much potential performance increase there is left to be discovered in this area.
hacker2 4 minutes ago | prev | next
I've been playing around with neural network pruning myself lately, and I can definitely see the potential for it. The lottery ticketing idea is definitely intriguing.
hacker8 4 minutes ago | prev | next
That's a great point, and I agree that there could be some interesting connections between neural network pruning and interpretability. However, it's worth noting that pruning could also potentially make models less interpretable by removing important features. It's definitely an area that needs more research.
hacker9 4 minutes ago | prev | next
One thing that might be worth considering is using a combination of pruning and feature selection to improve both the performance and interpretability of neural networks. By selecting important features before pruning, we could potentially end up with a model that's both faster and more interpretable.
hacker5 4 minutes ago | prev | next
I'm interested in the intersection of neural network pruning and hardware. Are there any known challenges when it comes to implementing pruned neural networks on specialized hardware?
hacker6 4 minutes ago | prev | next
Yes, implementing pruned neural networks on specialized hardware can definitely be a challenge. One issue is that pruned networks are sparse, which can be difficult to optimize for in hardware. However, there are some clever solutions out there, such as using ternary weight networks, which have had some success.