SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size · HackerPapers discussions

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size — mentioned by kajecounterhack

Tue, 14 Aug 2018 15:49:30 GMT

Inference speed and net size is definitely considered important and some researchers are definitely tackling it (e.g https://arxiv.org/abs/1602.07360). However it's more of an engineering area (meaning lots of engineers are working on it) and less of a research area (meaning fewer papers get produced per work) because there's not much in the way of performance-per-gflop benchmarks or whatever. CPU performance is good-ish when you're doing sequential stuff e.g it's not bad at old-school RNNs, but for image recognition using a cheap graphics unit is still going to be better, and if the question…

Paper and discussion history. A mention is not an endorsement.

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size — mentioned by vanderZwan

Wed, 13 Dec 2017 04:07:30 GMT

That's no different from a real-life network. Biology tends to solve this by growing connections as well as pruning them. Another comment mentioned Song Han's work into NN compression. Well, I looked up a recent paper and look what it says: > We discovered an interesting byproduct of model compression: re-densifying and retraining from a sparse model can improve the accuracy. That is, compared to a dense CNN baseline, dense → sparse → dense (DSD) training yielded higher accuracy > We now explain our DSD training strategy. On top of the sparse SqueezeNet (pruned 3x), we let the killed weights…

Paper and discussion history. A mention is not an endorsement.

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size — HN submission

Wed, 01 Mar 2017 12:46:23 GMT

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters

3 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.