# US 2021/0081785 A1 · Information processing device and method, and recording medium storing information processing program

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0081785 A1
- Document ID: US20210081785A1
- Status: Published application
- Kind: A1
- Application number: 17/002,814
- Filed: 2020-08-26
- Published: 2021-03-18
- Earliest claimed: 2019-09-13
- Assignees: Fujitsu Limited
- Applicants: Fujitsu Limited
- Inventors: Yasufumi Sakai
- Web page: https://patentgrep.com/patent/US20210081785A1

## CPC

- G06N 3/08: Learning methods
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/0495: Quantised networks; Sparse networks; Compressed networks
- G06N 3/063: using electronic means
- G06N 3/09: Supervised learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

An information processing device, includes a memory; and a processor coupled to the memory and configured to: determine a plurality of bit ranges after quantization for at least one of a plurality of types of variables to be used in a neural network, calculate a plurality of recognition rates of the neural network by using each of a plurality of variable groups which includes the plurality of types of variables, and in which a bit range of at least one of the plurality of types of variables is different, and determine to use a variable group of the plurality of variable groups, the variable group having a maximum recognition rate among the plurality of calculated recognition rates, for calculation of the neural network.

## Claims

### Claim 1

An information processing device, comprising:
a memory; and
a processor coupled to the memory and configured to:
determine a plurality of bit ranges after quantization for at least one of a plurality of types of variables to be used in a neural network,
calculate a plurality of recognition rates of the neural network by using each of a plurality of variable groups which includes the plurality of types of variables, and in which a bit range of at least one of the plurality of types of variables is different, and
determine to use a variable group of the plurality of variable groups, the variable group having a maximum recognition rate among the plurality of calculated recognition rates, for calculation of the neural network.

### Claim 2 (depends on 1)

The information processing device according to claim 1, wherein the processor is configured to:
execute the calculation of the neural network in such a manner that the calculation is executed in a plurality of calculation cycles each of which includes a group determination period and a calculation execution period,
calculate the recognition rates and determine the variable group having the maximum recognition rate are operated in the group determination period, and
in each of the plurality of calculation cycles, execute the calculation in the calculation execution period by using the variable group determined in the group determination period.

### Claim 3 (depends on 1)

The information processing device according to claim 1, wherein the processor is configured to determine, based on distribution of most significant bits when a determination target variable of the plurality of bit ranges is represented by a fixed-point number, the plurality of bit ranges from the most significant bit side of the distribution.

### Claim 4 (depends on 3)

The information processing device according to claim 3, wherein the processor is configured to:
calculate quantization errors when the determination target variable of the plurality of bit ranges is quantized in the plurality of bit ranges, and
determine the plurality of bit ranges from the most significant bit side of the distribution in ascending order of the calculated quantization errors.

### Claim 5 (depends on 3)

The information processing device according to claim 3, wherein the processor is configured to execute learning of the neural network by using the determined variable group having the maximum recognition rate.

### Claim 6 (depends on 3)

The information processing device according to claim 3, wherein the processor is configured to execute inference of the neural network by using the determined variable group having the maximum recognition rate.

### Claim 7

An information processing method for causing a processor included in an information processing device to execute a process, the process comprising:
determining a plurality of bit ranges after quantization for at least one of a plurality of types of variables to be used in a neural network;
calculating a plurality of recognition rates of the neural network by using each of a plurality of variable groups which includes the plurality of types of variables, and in which a bit range of at least one of the plurality of types of variables is different; and
determining to use a variable group of the plurality of variable groups, the variable group having a maximum recognition rate among the plurality of calculated recognition rates, for calculation of the neural network.

### Claim 8

A non-transitory computer-readable storage medium storing a program that causes a computer to execute a process, the process comprising:
determining a plurality of bit ranges after quantization for at least one of a plurality of types of variables to be used in a neural network;
calculating a plurality of recognition rates of the neural network by using each of a plurality of variable groups which includes the plurality of types of variables, and in which a bit range of at least one of the plurality of types of variables is different; and
determining to use a variable group of the plurality of variable groups, the variable group having a maximum recognition rate among the plurality of calculated recognition rates, for calculation of the neural network.
