Machine learning with model filtering and model mixing for edge devices in a heterogeneous environment

Patent metadata

Field	Value
Publication number	US-2016217387-A1
Application number	US-201514602843-A
Country	US
Kind code	A1
Filing date	Jan 22, 2015
Priority date	Jan 22, 2015
Publication date	Jul 28, 2016
Grant date	—

Abstract

Machine learning with model filtering and model mixing for edge devices in a heterogeneous environment is disclosed. In an example embodiment, an edge device includes a communication module, a data collection device, a memory, a machine learning module, and a model mixing module. The edge device analyzes collected data with a model for a first task, outputs a result, and updates the model to create a local model. The edge device communicates with other edge devices in a heterogeneous group, transmits a request for local models to the heterogeneous group, and receives local models from the heterogeneous group. The edge device filters the local models by structure metadata, including second local models, which relate to a second task. The edge device performs a mix operation of the second local models to generate a mixed model which relates to the second task, and transmits the mixed model to the heterogeneous group.

First claim

The invention is claimed as follows:

1 . An edge device comprising:
a communication module configured to communicate with a plurality of different edge devices; a data collection device configured to collect a first type of data; a memory configured to store data collected by the data collection device; a machine learning module; a model mixing module, wherein the edge device is configured to:
analyze, using a first model relating to a first predefined task, first data collected by the data collection device;
output a result including at least one of a prediction, a classification, a clustering, an anomaly detection, and a recognition;
update, based on a correctness of the result, the first model to create a first local model which relates to the first predefined task;
communicate with at least one other edge device in a heterogeneous group of edge devices, wherein the heterogeneous group of edge devices includes at least a first edge device and a second edge device, and the first edge device collects and analyzes the first type of data and the second edge device collects and analyzes a different second type of data;
transmit a request for local models to the heterogeneous group of edge devices;
receive a first plurality of local models from the heterogeneous group of edge devices;
filter the first plurality of local models by structure metadata, wherein the first plurality of local models includes a second plurality of local models, each of which relates to a second predefined task;
perform a mix operation of the second plurality of local models to generate a mixed model which relates to the second predefined task; and
transmit the mixed model to the heterogeneous group of edge devices.

2 . The edge device of claim 1 , wherein the first type of data is video image data and the second type of data is velocity data.

3 . The edge device of claim 1 , wherein the edge device is further configured to filter the plurality of local models by context metadata.

4 . The edge device of claim 1 , wherein the edge device is further configured to filter the plurality of local models by data distribution.

5 . The edge device of claim 1 , wherein the edge device is further configured to cluster the plurality of local models.

6 . The edge device of claim 1 , wherein the mix operation is determined using a size of a group of the plurality of local models and a computer power of the edge device.

7 . The edge device of claim 1 , wherein the mix operation is an averaging operation.

8 . The edge device of claim 1 , wherein the mix operation is a genetic algorithm operation.

9 . The edge device of claim 1 , wherein the mix operation is an enumeration operation.

10 . The edge device of claim 1 , wherein the mix operation is an ensemble operation.

11 . The edge device of claim 1 , wherein the first predefined task is the same task as the second predefined task.

12 . The edge device of claim 1 , wherein the edge device is further configured to replace the first local model with the mixed model.

13 . The edge device of claim 1 , wherein the first plurality of local models includes a third plurality of local models which is different from the second plurality of local models, and wherein the edge device is further configured to perform a mix operation of the third plurality of local models to generate a second mixed model which relates to a third predefined task which is different from the second predefined task.

14 . The edge device of claim 1 , wherein the edge device is one of a shopping cart device and a surveillance camera.

15 . The edge device of claim 1 , wherein the edge device is incorporated in an automobile.

16 . The edge device of claim 1 , wherein the edge device is an automatic teller machine.

17 . A computer readable medium storing instructions which, when executed by an edge device which comprises a communication module configured to communicate with a plurality of different edge devices, a data collection device configured to collect a first type of data, a memory configured to store data collected by the data collection device, a machine learning module, and a model mixing module, cause the edge device to:
analyze, using a first model relating to a first predefined task, first data collected by the data collection device; output a result including at least one of a prediction, a classification, a clustering, an anomaly detection, and a recognition; update, based on a correctness of the result, the first model to create a first local model which relates to the first predefined task; communicate with at least one other edge device in a heterogeneous group of edge devices, wherein the heterogeneous group of edge devices includes at least a first edge device and a second edge device, and the first edge device collects and analyzes the first type of data and the second edge device collects and analyzes a different second type of data; transmit a request for local models to the heterogeneous group of edge devices; receive a first plurality of local models from the heterogeneous group of edge devices; filter the first plurality of local models by structure metadata, wherein the first plurality of local models includes a second plurality of local models, each of which relates to a second predefined task; perform a mix operation of the second plurality of local models to generate a mixed model which relates to the second predefined task; and transmit the mixed model to the heterogeneous group of edge devices.

18 . A method comprising:
communicating, by a communication module in an edge device, with a plurality of different edge devices; collecting, with a data collection device, a first type of data; storing, in a memory, data collected by the data collection device; analyzing, using a first model relating to a first predefined task, first data collected by the data collection device; outputting a result including at least one of a prediction, a classification, a clustering, an anomaly detection, and a recognition; updating, based on a correctness of the result, the first model to create a first local model which relates to the first predefined task; communicating with at least one other edge device in a heterogeneous group of edge devices, wherein the heterogeneous group of edge devices includes at least a first edge device and a second edge device, and the first edge device collects and analyzes the first type of data and the second edge device collects and analyzes a different second type of data; transmitting a request for local models to the heterogeneous group of edge devices; receiving a first plurality of local models from the heterogeneous group of edge devices; filtering the first plurality of local models by structure metadata, wherein the first plurality of local models includes a second plurality of local models, each of which relates to a second predefined task; performing a mix operation of the second plurality of local models to generate a mixed model which relates to the second predefined task; and transmitting the mixed model to the heterogeneous group of edge devices.

19 . The method of claim 18 , further comprising filtering the plurality of local models by context metadata.

20 . The method of claim 18 , further comprising filtering the plurality of local models by data distribution.

Assignees

• Preferred Networks Inc

Inventors

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

Classifications

• G06N7/005

  Physics · mapped topic

• G06F9/4881

  Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues · CPC title

• G06N99/005 Primary

  Physics · mapped topic

• G06N20/00 Primary

  Machine learning · CPC title