Processing for multiple input data sets

Patent metadata

Field	Value
Publication number	US-11475306-B2
Application number	US-201815933201-A
Country	US
Kind code	B2
Filing date	Mar 22, 2018
Priority date	Mar 22, 2018
Publication date	Oct 18, 2022
Grant date	Oct 18, 2022

Abstract

Disclosed herein are techniques for performing multi-layer neural network processing for multiple contexts. In one embodiment, a computing engine is set in a first configuration to implement a second layer of a neural network and to process first data related to a first context to generate first context second layer output. The computing engine can be switched from the first configuration to a second configuration to implement a first layer of the neural network. The computing engine can be used to process second data related to a second context to generate second context first layer output. The computing engine can be set to a third configuration to implement a third layer of the neural network to process the first context second layer output and the second context first layer output to generate a first processing result of the first context and a second processing result of the second context.

First claim

What is claimed is:

1. A method of processing data, comprising:
setting a computing engine in a first configuration to implement a first layer of a neural network;
processing first data related to a first context using the computing engine in the first configuration to generate a first layer output of the first context;
setting the computing engine in a second configuration to implement a second layer of the neural network;
processing the first layer output of the first context using the computing engine in the second configuration to generate a second layer output of the first context;
switching the computing engine from the second configuration back to the first configuration;
processing second data related to a second context using the computing engine in the first configuration to generate a first layer output of the second context;
setting the computing engine back to the second configuration to implement the second layer of the neural network;
processing the first layer output of the second context in the second configuration to generate a second layer output of the second context;
setting the computing engine in a third configuration to implement a third layer of the neural network; and
processing the second layer output of the first context and the second layer output of the second context using the computing engine in the third configuration to generate a third layer output of the first context and a third layer output of the second context.

2. The method of claim 1 , further comprising:
processing third data related to a third context using the computing engine in the first configuration to generate a first layer output of the third context; and
processing the first layer output of the third context and the first layer output of the first context using the computing engine in the second configuration to generate, respectively, a second layer output of the third context and the second layer output of the first context.

3. The method of claim 2 , further comprising:
storing the first layer output of the first context and the first layer output of the third context in a memory device;
configuring a first portion of the computing engine to receive the first layer output of the first context from the memory device as input;
configuring a second portion of the computing engine to receive the first layer output of the third context from the memory device as input;
processing the first layer output of the first context using the first portion of the computing engine; and
processing the first layer output of the third context using the second portion of the computing engine.

4. The method of claim 3 , wherein the processing of the first layer output of the first context using the first portion of the computing engine and the processing of the first layer output of the third context using the second portion of the computing engine are performed in parallel.

5. The method of claim 1 , further comprising:
processing fourth data related to a fourth context using the computing engine in the first configuration to generate a first layer output of the fourth context; and
processing the first layer output of the second context and the first layer output of the fourth context using the computing engine in the second configuration to generate, respectively, the second layer output of the second context and a second layer output of the fourth context.

6. The method of claim 5 , further comprising:
storing the first layer output of the second context and the first layer output of the fourth context in a memory device;
configuring a first portion of the computing engine to receive the first layer output of the second context from the memory device as input;
configuring a second portion of the computing engine to receive the first layer output of the fourth context from the memory device as input;
processing the first layer output of the second context using the first portion of the computing engine; and
processing the first layer output of the fourth context using the second portion of the computing engine.

7. The method of claim 6 , wherein the processing of the first layer output of the second context using the first portion of the computing engine and the processing of the first layer output of the fourth context using the second portion of the computing engine are performed substantially in parallel.

8. The method of claim 1 , further comprising:
storing, at a memory device, the second layer output of the first context;
storing, at the memory device in addition to the second layer output of the first context, the second data related to the second context; and
storing, at the memory device in addition to the second layer output of the first context and the second data related to the second context, the first layer output of the second context.

9. The method of claim 8 , further comprising:
overwriting at least a part of the second data related to the second context or the first layer output of the second context stored at the memory device with the second layer output of the second context.

10. The method of claim 9 , further comprising:
configuring a third portion of the computing engine to receive the second layer output of the first context from the memory device as input;
configuring a fourth portion of the computing engine to receive the second layer output of the second context from the memory device as input;
processing the second layer output of the first context using the third portion of the computing engine; and
processing the second layer output of the second context using the fourth portion of the computing engine.

11. The method of claim 10 , wherein the processing of the second layer output of the first context using the third portion of the computing engine and the processing of the second layer output of the second context using the fourth portion of the computing engine are performed substantially in parallel.

12. The method of claim 1 , wherein processing the first data related to the first context using the computing engine in the first configuration to generate the first layer output of the first context comprises:
performing one or more convolution computations between the first data and a set of weights associated with the first layer of the neural network.

13. The method of claim 12 , further comprising:
processing results of the one or more convolution computations by an activation function engine to generate the first layer output of the first context.

Assignees

• Amazon Tech Inc

Inventors

• Vantrease Dana Michelle
• Diamant Ron
• Volpe Thomas A
• Huang Randy

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06N3/082 Primary

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06F3/0673

  Single storage device · CPC title