Lecture 8: Subgradient method (continued); Proximal gradient descent and acceleration

preview_player
Показать описание

Рекомендации по теме
Комментарии
Автор

@37:00 It's interesting that in Deep Learning we usually use (minibatch) /randomized/ SGD, but without replacement to make sure we don't skip any examples. This looks like a cyclic randomized rule!

kiuhnmmnhuik
Автор

46:52 proximal gradient method

thanks

choungyoungjae
Автор

@20:00 I don't understand the doubts here. If we use Polyak's step sizes then the subgradient method *must* converge. Nobody said that the subgradients must also converge.

kiuhnmmnhuik
Автор

Hey Ryan, Thank you very much for uploading the lecture it is really comprehensive. Is there a way i can get the slides? I'm currently writing a bachelor thesis on dictionary learning.

Cheers

CoinedBeatz
Автор

The last part was quite painful to watch because of the blur.

kiuhnmmnhuik