Machine learning heterogeneous edge device, method, and system

Patent metadata

Field	Value
Publication number	US-2018253665-A1
Application number	US-201815970294-A
Country	US
Kind code	A1
Filing date	May 3, 2018
Priority date	Jan 22, 2015
Publication date	Sep 6, 2018
Grant date	—

Abstract

A machine learning heterogeneous edge device, method, and system are disclosed. In an example embodiment, an edge device includes a communication module, a data collection device, a memory, a machine learning module, a group determination module, and a leader election module. The edge device analyzes collected data with a model, outputs a result, and updates the model to create a local model. The edge device communicates with other edge devices in a heterogeneous group. The edge device determines group membership and determines a leader edge device. The edge device receives a request for the local model, transmits the local model to the leader edge device, receives a mixed model created by the leader edge device performing a mix operation of the local model and a different local model, and replaces the local model with the mixed model.

First claim

1 - 20 . (canceled)

21 . An edge device comprising:
a memory; and a processor coupled to the memory and configured to:
collect data and store the data in the memory,
analyze, using a local model, the data collected,
transmit requests for group information to a plurality of edge devices, the group information being information for forming a group,
receive the group information from the plurality of edge devices; and
form the group and determine a leader edge device based on the group information,
wherein, when the edge device is determined to be the leader edge device, the processor performs:
transmitting requests for local models to edge devices in the group,
receiving local models from the edge devices in the group,
generating a mixed model by mixing models from the local models; and
transmitting the mixed model to the edge devices in the group.

22 . The edge device of claim 21 , wherein the determination of the leader edge device is based on at least one of a user designation, environmental conditions, a random assignment, an ID based selection, a consensus protocol, and a function-based selection.

23 . The edge device of claim 21 , wherein the group is determined based on a geographic proximity of the plurality of edge devices.

24 . The edge device of claim 21 , wherein the leader edge device filters the local models to select the models for generating the mixed model.

25 . The edge device of claim 21 further configured to update the local model of the edge device based on analysis results by the edge device itself.

26 . The edge device of claim 21 , wherein the leader edge device generates the mixed model also with the local model of the leader edge device.

27 . The edge device of claim 21 , wherein the plurality of edge devices comprise:
a first edge device analyzing data collected by the first edge device; and a second edge device analyzing data collected by the second edge device, wherein the data collected by the second edge device are different types of data from the data collected by the first edge device.

28 . The edge device of claim 21 , wherein the local models are updated from a global model different from any of the local models.

29 . The edge device of claim 21 further configured to replace, by the processor, the local model of the edge device with the mixed model.

30 . A method comprising:
collecting data and storing the data in a memory in an edge device, analyzing, by a processor in the edge device, using a local model, the data collected, transmitting, by the processor, requests for group information to a plurality of edge devices, the group information being information for forming a group, receiving, by the processor, the group information from the plurality of edge devices; and forming, by the processor, the group and determining a leader edge device based on the group information, wherein the method further comprises: transmitting, by a processor in the leader edge device, requests for local models to edge devices in the group, receiving, by the processor in the leader edge device, local models from the edge devices in the group, generating, by the processor in the leader edge device, a mixed model by mixing models from the local models; and transmitting, by the processor in the leader edge device, the mixed model to the edge devices in the group.

31 . The method of claim 30 , wherein the determining the leader edge device is performed based on at least one of a user designation, environmental conditions, a random assignment, an ID based selection, a consensus protocol, and a function-based selection.

32 . The method of claim 30 further comprising determining, by the processor, the group based on a geographic proximity of the plurality of edge devices.

33 . The method of claim 30 further comprising filtering, by the processor in the leader edge device, the local models to select the models for generating the mixed model.

34 . The method of claim 30 further comprising updating, by the processor, the local model of the edge device based on analysis results by the edge device itself.

35 . The method of claim 30 , wherein generating the mixed model also performed with the local model of the leader edge device.

36 . The method of claim 30 , wherein the plurality of edge devices comprise:
a first edge device analyzing data collected by the first edge device, a second edge device analyzing data collected by the second edge device, wherein the data collected by the second edge device are different types of data from the data collected by the first edge device.

37 . The method of claim 30 further comprising updating, by the processor, the local model from a global model different from any of the local models.

38 . The method of claim 30 further comprising replacing, by the processor, the local model of the edge device with the mixed model.

39 . A non-transitory computer readable medium storing a program causing an edge device to:
collect data and store the data in a memory in the edge device; analyze, using a local model, the data collected; transmit requests for group information to a plurality of edge devices, the group information being information for forming a group; receive the group information from the plurality of edge devices; and form the group and determine a leader edge device based on the group information; wherein the program further causes the leader edge device to: transmit requests for local models to edge devices in the group; receive local models from the edge devices in the group; generate a mixed model by mixing models from the local models; and transmit the mixed model to the edge devices in the group.

40 . The non-transitory computer readable medium of claim 39 storing the program further causing the edge device to update the local model of the edge device based on analysis results by the edge device itself.

Assignees

• Preferred Networks Inc

Inventors

• Okanohara Daisuke
• CLAYTON Justin
• Nishikawa Toru
• Hido Shohei
• Kubota Nobuyuki
• Ota Nobuyuki
• Tokui Seiya

Classifications

• G06N99/005 Primary

  Physics · mapped topic

• G06N20/00 Primary

  Machine learning · CPC title