ECE 804 - Dr Maya Gupta -Stein Paradox and Multi-task Averaging
Автор: NC State ECE
Загружено: 2015-06-17
Просмотров: 1315
Описание: In the 1960's, Stein showed that you could make better estimates of the means of different, independent random variables if you mixed their samples together. That is, somewhat surprisingly, you can do a better job (in total mean-squared error sense) of estimating the price of tea in China andthe probability of rain in Spain if you mix both data-sets (appropriately). We re-frame this problem of jointly estimating multiple means as a multi-task learning problem, and apply a standard machine learning structural risk minimization framework. This formulation may be more intuitive and extensible than Stein's approach, and it produces nice theoretical properties. We'll show improved practical performance over James-Stein estimation for estimating students' grades, Amazon customer reviews, customer purchases, and the length of kings' reigns. We'll discuss large-scale optimization strategies for scaling to "big data." This multi-task averaging strategy may be applicable for any algorithm or application where multiple means are estimated.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
