Papers of the day   All papers

ONCE-FOR-ALL: TRAIN ONE NETWORK AND SPECIALIZE IT FOR EFFICIENT DEPLOYMENT

Comments

PyTorch: The Once-For-All (OFA) network from Han Cai, @SongHan_MIT et. al. Train one flexible network, deploy subsets that specialize to Mobile, Cloud, IOT efficiently, without retraining! Code: https://github.com/mit-han-lab/once-for-all Paper: https://arxiv.org/abs/1908.09791 News: http://news.mit.edu/2020/artificial-intelligence-ai-carbon-footprint-0423 https://t.co/MegrXNwBUA

2 replies, 325 likes


Chris Albertyn: #AI has the capacity to optimise our #sustainability practices...but let's not forget it has an intrinsic #carbonfootprint (i.e. https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/). Will the following be a solution? https://singularityhub.com/2020/05/04/this-once-for-all-neural-network-could-slash-ais-carbon-footprint/ (pre-print: https://arxiv.org/abs/1908.09791) @EricTopol @NHSX @HDR_UK https://t.co/7p7JOvumwf

0 replies, 2 likes


Dr William Marshall: Paper: Once-for-All: Train One Network and Specialise it for Efficient Deployment. MIT researchers have created a method of reducing the power requirement - and hence carbon footprint - of #AI system development. #DNN https://arxiv.org/pdf/1908.09791.pdf

0 replies, 1 likes


Content

Found on Apr 29 2020 at https://arxiv.org/pdf/1908.09791.pdf

PDF content of a computer science paper: ONCE-FOR-ALL: TRAIN ONE NETWORK AND SPECIALIZE IT FOR EFFICIENT DEPLOYMENT