Dynamically modifying a boundary of a deep learning network

Patent metadata

Field	Value
Publication number	US-10679140-B2
Application number	US-201414506972-A
Country	US
Kind code	B2
Filing date	Oct 6, 2014
Priority date	Oct 6, 2014
Publication date	Jun 9, 2020
Grant date	Jun 9, 2020

Abstract

A connection between a user device and a network server is established. Via the connection, a deep learning network is formed for a processing task. A first portion of the deep learning network operates on the user device and a second portion of the deep learning network operates on the network server. Based on cooperation between the user device and the network server, a boundary between the first portion and the second portion of the deep learning network is dynamically modified based on a change in a performance indicator that could affect the processing task.

First claim

What is claimed is:

1. A method comprising:
establishing a connection between a user device and a network server;
forming, via the connection, a deep learning network for a processing task that involves performing recognition on data that is streaming from the user device via a first process, a first portion of the deep learning network operating on the user device and a second portion of the deep learning network operating on the network server, the first and second portions each processing a different level of representation of the processing task;
based on cooperation between the user device and the network server while the streaming of the data is ongoing and without pausing the processing task, dynamically modifying a boundary between the first portion and the second portion of the deep learning network based on a change of a performance indicator that could affect the processing task, the modification of the boundary changing the respective different levels of representation performed by the first and second portions of the deep learning network; and
continuing the processing task via a second process that operates in parallel with the first processes, the second process utilizing the modified boundary to perform recognition on the data.

2. The method of claim 1 , wherein the boundary is modified after the processing task is complete for the benefit of subsequent similar processing tasks.

3. The method of claim 1 , wherein the boundary comprises an interface between hierarchical levels of the deep learning network.

4. The method of claim 1 , wherein dynamically modifying the boundary comprises signaling between the user device and the network server to modify a current session.

5. The method of claim 1 , wherein dynamically modifying the boundary comprises selecting different ones of multiple application program interfaces available between the user device and the network server.

6. The method of claim 1 , wherein the processing task is performed on behalf of a user of the user device.

7. The method of claim 1 , wherein the user device comprises a mobile device.

8. A network server, comprising:
a network interface;
a processor coupled to the network interface, the processor configured to:
establish a connection with a user device;
form, via the connection, a deep learning network for a processing task on behalf of the user device, the processing task performing recognition on data that is streaming from the user device via a first process, wherein a first portion of the deep learning network operates on the user device and a second portion of the deep learning network operates on the network server, the first and second portions each processing a different level of representation of the processing task;
in cooperation with the user device, dynamically modify a boundary between the first portion and the second portion of the deep learning network while the streaming of the data is ongoing and without pausing the processing task, the modification of the boundary based on a change in a performance indicator that could affect the processing task, the modification of the boundary changing the respective different levels of representation performed by the first and second portions of the deep learning network; and
continuing the processing task via a second process that operates in parallel with the first processes, the second process utilizing the modified boundary to perform recognition on the data.

9. The network server of claim 8 , wherein the boundary comprises an interface between hierarchical levels of the deep learning network.

10. The network server of claim 8 , wherein the boundary is dynamically modified based on signaling between the user device and the network server.

11. The network server of claim 8 , wherein the boundary is dynamically modified based on selection of one of multiple interfaces available between the user device and the network server.

12. A user device, comprising:
a sensor;
a processor coupled to the sensor, the processor configured to:
encode signals from the sensor into a data stream;
send the data stream to a neural network for a processing task that involves a recognition of the data stream via a first process, wherein a first portion of the neural network operates on the user device and a second portion of the neural network operates on a network server in communication with the user device, the first and second portions each processing a different level of representation of the processing task;
in cooperation with the network server while the streaming of the data is ongoing and without pausing the processing task, dynamically modify a boundary between the first portion and the second portion of the neural network based on a change in a performance indicator that could affect the processing task, the modification of the boundary changing the respective different levels of representation performed by the first and second portions of the neural network and
continuing the processing task via a second process that operates in parallel with the first processes, the second process utilizing the modified boundary to perform recognition on the data.

13. The user device of claim 12 , wherein the boundary comprises an interface between hierarchical levels of a deep-belief network.

14. The user device of claim 12 , wherein the boundary is dynamically modified based on signaling between the user device and the network server.

15. The user device of claim 12 , wherein the boundary is dynamically modified based on selection of one of multiple interfaces available between the user device and the network server.

16. The user device of claim 12 , wherein the user device comprises a mobile device.

17. The method of claim 1 , wherein the processing task involves at least one of speech recognition and image recognition.

18. The method of claim 1 , further comprising changing a resolution of the data streaming from the user device based on the change of the performance indicator.

Assignees

• Seagate Technology Llc

Inventors

• Gomez Kevin Arthur
• Dropps Frank
• Goss Ryan James
• Trantham Jon
• Khoueir Antoine

Classifications

• G06N7/005

  Physics · mapped topic

• H04L67/10

  in which an application is distributed across nodes in the network (software deployment G06F8/60; multiprogramming arrangements G06F9/46) · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06N3/0454

  Physics · mapped topic

• H04L67/141

  Setup of application sessions (admission control or resource allocation in data switching networks H04L47/70) · CPC title