公开(公告)号：US20190303406A1

公开(公告)日：2019-10-03

申请号：US16383360

申请日：2019-04-12

Applicant: Apple Inc.

Inventor： Haijie Gu ,  Yucheng Low ,  Carlos Guestrin

IPC: G06F16/901 ,  G06F16/958 ,  G06F16/25 ,  G06F16/2453

Abstract: A method of optimizing graph operations is performed by a computing system. The method comprises: (1) receiving a first request to perform a first operation on a first graph, where the first graph comprises a set of vertices and a set of edges, each edge connecting a pair of vertices, and each vertex having one or more associated properties; (2) logging the first request, but not performing the first operation; (3) receiving a second request to perform a second operation; (4) logging the second request, but not performing the second operation; (5) receiving a query for data from the first graph, where the data includes property values for one or more vertices; (6) in response to the query: (a) generating a second graph by optimizing and performing the first and second operations; and (b) returning data responsive to the query, where the returned data is based on the second graph.

公开(公告)号：US10331740B2

公开(公告)日：2019-06-25

申请号：US14619020

申请日：2015-02-10

Applicant: Apple Inc.

Inventor： Yucheng Low ,  Haijie Gu ,  Ping Wang ,  Evan Samanas ,  Sethu Raman ,  Carlos Guestrin

IPC: G06F17/30 ,  G06F16/901 ,  G06F16/2453 ,  G06F16/25 ,  G06F16/958 ,  G06F16/22 ,  G06F16/23 ,  G06F16/2455 ,  G06F16/27 ,  G06F16/21

Abstract: A method receives a first request from a client object at a device. The first request specifies a data source. In response to the first request, the method uploads data from the data source, stores the data as a plurality of first columns, and instantiates a first server object that provides access to the first columns. The method later receives a second request from the client object. The second request specifies a transformation of the data. In response to the second request, the method stores one or more additional columns and instantiates a second server object that provides access to the additional columns and one or more of the first columns. Each of the additional columns is constructed from the first columns according to the requested transformation, and each of the additional columns includes a plurality of data values all having the same data type.

公开(公告)号：US10262078B2

公开(公告)日：2019-04-16

申请号：US14619025

申请日：2015-02-10

Applicant: Apple Inc.

Inventor： Haijie Gu ,  Yucheng Low ,  Carlos Guestrin

IPC: G06F17/30

Abstract: A method of optimizing graph operations is performed by a computing system. The method comprises: (1) receiving a first request to perform a first operation on a first graph, where the first graph comprises a set of vertices and a set of edges, each edge connecting a pair of vertices, and each vertex having one or more associated properties; (2) logging the first request, but not performing the first operation; (3) receiving a second request to perform a second operation; (4) logging the second request, but not performing the second operation; (5) receiving a query for data from the first graph, where the data includes property values for one or more vertices; (6) in response to the query: (a) generating a second graph by optimizing and performing the first and second operations; and (b) returning data responsive to the query, where the returned data is based on the second graph.

公开(公告)号：US20210089887A1

公开(公告)日：2021-03-25

申请号：US16832934

申请日：2020-03-27

Applicant: Apple Inc.

Inventor： Tyler B. Johnson ,  Carlos E. Guestrin ,  Pulkit Agrawal ,  Haijie Gu

IPC: G06N3/08

Abstract: A method includes determining a training scale for training a machine-learning model, defining a group of worker nodes having a number of worker nodes that is selected according to the training scale, and determining an average gradient of a loss function during a training iteration using the group of worker nodes. The method also includes determining a variance value for the average gradient of the loss function, determining a gain ratio based on the variance value for the average gradient of the loss function, and determining a learning rate parameter based on a learning rate schedule and the gain ratio. The method also includes determining updated parameters for the machine-learning model using the learning rate parameter and the average gradient of the loss function.