Computer-readable recording medium having stored therein learning program, learning method, and learning apparatus

Patent metadata

Field	Value
Publication number	US-11429863-B2
Application number	US-201916670322-A
Country	US
Kind code	B2
Filing date	Oct 31, 2019
Priority date	Nov 12, 2018
Publication date	Aug 30, 2022
Grant date	Aug 30, 2022

Abstract

A learning method includes: acquiring input data and correct answer information, the input data including a set of multiple pieces of relationship data in which relationships between variables are recorded respectively; determining conversion rule corresponding to each of the multiple pieces of relationship data such that relationships before and after a conversion of a common variable commonly in the multiple pieces of relationship data are the same, when converting a variable value in each of the multiple pieces of relationship data into converted data rearranging the variable values in an order of input; converting each of the multiple pieces of relationship data into a multiple pieces of the converted data according to each corresponding conversion rule; and inputting a set of the multiple pieces of converted data to the neural network and causing the neural network to learn a learning model based on the correct answer information.

First claim

What is claimed is:

1. A non-transitory computer-readable recording medium having stored therein a learning program causing a computer to execute processing, the processing comprising:
acquiring input data and correct answer information added to the input data, the input data including a set of a plurality of pieces of relationship data in which relationships between variables are recorded respectively;
determining each conversion rule corresponding to each of the plurality of pieces of relationship data such that correspondence relationships before and after a conversion of a common variable commonly included in the plurality of pieces of relationship data are the same, when converting a variable value included in each of the plurality of pieces of relationship data into converted data rearranging the variable values in an order of input, according to each collation pattern in which a reference for ordering the variable value which is input to a neural network and which corresponds to each of the plurality of pieces of relationship data, is defined by an array of a plurality of reference values;
converting each of the plurality of pieces of relationship data into a plurality of pieces of the converted data according to each corresponding conversion rule; and
inputting a set of the plurality of pieces of converted data to the neural network as the input data, thereby causing the neural network to learn a learning model based on the correct answer information.

2. The non-transitory computer-readable recording medium according to claim 1 ,
wherein the determining includes determining each conversion rule such that a total sum of each of similarities between each of the plurality of pieces of converted data generated from each of the plurality of pieces of relationship data and corresponding collation patterns is maximized, while satisfying a condition that the correspondence relationships before and after the conversion of the common variable are the same.

3. The non-transitory computer-readable recording medium according to claim 2 ,
wherein the determining includes calculating a similarity between each of the plurality of pieces of converted data generated from each of the plurality of pieces of relationship data and the corresponding collation patterns, multiplying each calculated similarity by each importance set for each of the plurality of pieces of relationship data, and determining each conversion rule such that a total sum of each multiplied value is maximized.

4. The non-transitory computer-readable recording medium according to claim 3 ,
wherein the executing learning includes
calculating an error between an output value, obtained by inputting variable values included in the input data to the neural network in which a plurality of input layers are provided for each relationship in an order of input, and the correct answer information, and
updating a parameter corresponding to each of the plurality of input layers by using a multiplication result of each of the plurality of pieces of converted data and the error.

5. The non-transitory computer-readable recording medium according to claim 4 ,
wherein the executing learning includes
acquiring a plurality of pieces of error information corresponding to each of the plurality of input layers by executing error back propagation based on the error,
determining each conversion rule such that a total sum of each of the similarities is maximized while satisfying the condition when each of the plurality of reference values included in each collation pattern is minutely changed,
generating the plurality of pieces of converted data according to each conversion rule,
calculating a change amount of each collation pattern based on a variation between the plurality of pieces of converted data before the minute change and the plurality of pieces of converted data after the minute change, and the plurality of pieces of error information, and
updating each collation pattern by using the change amount of each collation pattern.

6. The non-transitory computer-readable recording medium according to claim 5 ,
wherein the executing learning includes
determining each conversion rule such that, when each importance is minutely changed, the total sum of each of the similarities is maximized while satisfying the condition,
generating the plurality of pieces of converted data according to each conversion rule,
calculating a change amount of each importance based on the variation between the plurality of pieces of converted data before the minute change and the plurality of pieces of converted data after the minute change, and the plurality of pieces of error information, and
updating each importance by using the change amount of each importance.

7. A learning method executed by a computer, the method comprising:
acquiring input data and correct answer information added to the input data, the input data including a set of a plurality of pieces of relationship data in which relationships between variables are recorded respectively;
determining each conversion rule corresponding to each of the plurality of pieces of relationship data such that correspondence relationships before and after a conversion of a common variable commonly included in the plurality of pieces of relationship data are the same, when converting a variable value included in each of the plurality of pieces of relationship data into converted data rearranging the variable values in an order of input, according to each collation pattern in which a reference for ordering the variable value which is input to a neural network and which corresponds to each of the plurality of pieces of relationship data, is defined by an array of a plurality of reference values;
converting each of the plurality of pieces of relationship data into a plurality of pieces of the converted data according to each corresponding conversion rule; and
inputting a set of the plurality of pieces of converted data to the neural network as the input data, thereby causing the neural network to learn a learning model based on the correct answer information.

8. A learning apparatus comprising:
a memory;
a processor coupled to the memory, the processor being configured to
execute an acquisition processing that includes acquiring input data and correct answer information added to the input data, the input data including a set of a plurality of pieces of relationship data in which relationships between variables are recorded respectively;
execute a determining processing that includes determining each conversion rule corresponding to each of the plurality of pieces of relationship data such that correspondence relationships before and after a conversion of a common variable commonly included in the plurality of pieces of relationship data are the same, when converting a variable value included in each of the plurality of pieces of relationship data into converted data rearranging the variable values in an order of input, according to each collation pattern in which a reference for ordering the variable value which is input to a neural network and which corresponds to each of the plurality of pieces of relationship data, is defined by an array of a plurality of reference values;
execute a conversion processing that includes converting each of the plurality of pieces of relationship data into a plurality of pieces of the converted data according to each corresponding conversion rule; and
execute a learning processing that includes inputting a set of the plurality of pieces of converted data to the neural network as the input data, thereby causing the neural network to learn a learning model based on the correct answer information.

Assignees

• Fujitsu Ltd

Inventors

• Matsuo Tatsuru
• Maruhashi Koji

Classifications

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• G06K9/6256

  Physics · mapped topic