Exascale Deep Learning for Climate Analytics (TF Dev Summit ‘19)

preview_player
Показать описание
Climate change will have fundamental socio-economic impact and it is imperative for us to understand it better. This talk will show how TensorFlow was utilized on the world’s fastest supercomputer in order to extract pixel level segmentation masks of extreme weather phenomena in climate simulation data, thereby enabling climate scientists to perform high-fidelity, fine grained geo-spatial analyses of the effects of climate change.

Speaker: Thorsten Kurth, Lawrence Berkeley National Library

event: TensorFlow Dev Summit 2019; re_ty: Publish; product: TensorFlow - General; fullname: Thorsten Kurth;
Рекомендации по теме
Комментарии
Автор

12:00 A minor optimisation would be to run the network with no lag for a few iterations first, then use that to initialise the weights of the lag=1 network and continue training.

deep.space.
Автор

Will there be any video about 2.0 and porting models?

abrampers