Tomer Koren - Why does SGD generalize better than others do
Автор: HUJI Machine Learning Club
Загружено: 2021-07-28
Просмотров: 397
Описание:
Delivered on June 10, 2021
Speaker:
Tomer Koren, TAU
Title:
Why does SGD generalize (better than others do)?
Abstract:
Stochastic Gradient Descent (SGD) is the canonical method for training machine learning models. Yet, the question of “explaining why” it generalizes remains perplexing and receives continued attention in the literature, even though there are remarkably simple analyses of its generalization performance. In this talk, I will propose concrete ways to formalize the “why” question and give some (partial) answers, by exhibiting a sharp separation between SGD and closely related gradient methods. Time permitting, I will also discuss relations to algorithmic stability in optimization and to differential privacy.
Online Calendar:
Learning Club @ HUJI
https://www.google.com/calendar/embed...
Mailing List:
subscription manager: http://mailman.cs.huji.ac.il/mailman/...
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
