公开(公告)号：US20170236072A1

公开(公告)日：2017-08-17

申请号：US15429216

申请日：2017-02-10

Applicant: Google Inc.

Inventor： Steffen Rendle ,  Dennis Craig Fetterly ,  Eugene J. Shekita ,  Bor-yiing Su

IPC: G06N99/00 ,  G06N7/00 ,  G06F3/06

CPC classification number: G06N20/00 ,  G06F3/0644 ,  G06F3/0655 ,  G06F3/067 ,  G06F9/46 ,  G06N7/00

Abstract: The present disclosure provides a new scalable coordinate descent (SCD) algorithm and associated system for generalized linear models whose convergence behavior is always the same, regardless of how much SCD is scaled out and regardless of the computing environment. This makes SCD highly robust and enables it to scale to massive datasets on low-cost commodity servers. According to one aspect, by using a natural partitioning of parameters into blocks, updates can be performed in parallel a block at a time without compromising convergence. Experimental results on a real advertising dataset are used to demonstrate SCD's cost effectiveness and scalability.

公开(公告)号：US10482392B2

公开(公告)日：2019-11-19

申请号：US15429216

申请日：2017-02-10

Applicant: Google Inc.

Inventor： Steffen Rendle ,  Dennis Craig Fetterly ,  Eugene J. Shekita ,  Bor-yiing Su

IPC: G06N20/00 ,  G06F3/06 ,  G06N7/00 ,  G06F9/46

Abstract: The present disclosure provides a new scalable coordinate descent (SCD) algorithm and associated system for generalized linear models whose convergence behavior is always the same, regardless of how much SCD is scaled out and regardless of the computing environment. This makes SCD highly robust and enables it to scale to massive datasets on low-cost commodity servers. According to one aspect, by using a natural partitioning of parameters into blocks, updates can be performed in parallel a block at a time without compromising convergence. Experimental results on a real advertising dataset are used to demonstrate SCD's cost effectiveness and scalability.