Data generation method and computer system

Patent metadata

Field	Value
Publication number	US-10783184-B2
Application number	US-201715593428-A
Country	US
Kind code	B2
Filing date	May 12, 2017
Priority date	Jun 30, 2016
Publication date	Sep 22, 2020
Grant date	Sep 22, 2020

Abstract

A computer includes a data generation unit and a storage unit which retains graph information for managing a graph configured from a plurality of vertexes and sides. The data generation unit performs acquiring a plurality of data and graph information and assuring storage regions in number equal to the number of vertexes, converting each data into an input value and setting at least one input value to a storage region corresponding to at least one vertex, executing an updating process for updating a value set to a storage region corresponding to a first vertex using the value set to the storage region corresponding to the first vertex and a value set to a storage region corresponding to a different vertex directly connected to the first vertex, and outputting a set of values set to the storage regions corresponding to the vertexes as the feature value.

First claim

What is claimed is:

1. A data generation method for a computer system which includes a plurality of computers each including a processor, a memory connected to the processor and a network interface connected to the processor,
at least one of the computers being programmed to acquire a plurality of time series data and generate pattern data representative of a feature value for identifying a class to which an identification target belongs using the plurality of time series data,
at least one of the computers including a memory configured to retain graph information for managing a graph configured from a plurality of vertexes and sides which connect the plurality of vertexes to each other,
the data generation method comprising:
acquiring the plurality of data and the graph information and assuring storage regions in number equal to the number of vertexes included in the graph for storing the plurality of data;
converting each of the plurality of time series data into an input value and setting at least one input value to a storage region corresponding to at least one of the vertexes included in the graph;
executing an updating process for updating a value set to a storage region corresponding to a first vertex using the value set to the storage region corresponding to the first vertex and a value set to a storage region corresponding to a different vertex directly connected to the first vertex; and
outputting a set of values set to the storage regions individually corresponding to the plurality of vertexes included in the graph as the pattern data, the pattern data being the feature value generated from the plurality of time series data and being given as the set of values set to the vertexes of the graph, the pattern data being indicative of a transient response and being the feature value on which an influence between the plurality of time series data and a transient of the time series data are reflected.

2. The data generation method according to claim 1 , wherein
the memory retains definition information of the vertexes to which the plurality of input values are individually set;
converting the data includes specifying a vertex to which each of the plurality of input values is to be set based on the definition information;
executing the updating process includes updating the value set to the storage region corresponding to the first vertex based on a reaction diffusion equation in which the value set to the storage region corresponding to the first vertex and the value set to the storage region corresponding to the different vertex connected to the first vertex are used; and
outputting the set of values includes outputting a set of values set to the storage regions individually corresponding to the plurality of vertexes included in the graph after the updating process is executed by a plural number of times as the pattern data.

3. The data generation method according to claim 2 , wherein
the storage regions corresponding to the vertexes have stored therein a number of particles which correspond to each of the plurality of data;
converting the data includes
converting each of the plurality of data into a number of particles corresponding to each of the plurality of data, and
setting the number of particles corresponding to each of the plurality of data to a storage region corresponding to at least one of the vertexes included in the graph; and

executing the updating process includes
deciding whether or not a particle set to the first vertex is to be moved based on the reaction diffusion equation;
determining a vertex of a movement destination of the particle set to the first vertex based on a result of the decision, and
updating the number of particles set to the storage region corresponding to the first vertex based on the determination.

4. The data generation method according to claim 3 , further comprising:
managing a plurality of particle storage regions for managing the number of particles corresponding to the plurality of data;
wherein the plurality of vertexes included in the graph include a deletion vertex in which the number of particles set to the storage region corresponding to the vertex is initialized after the updating process is executed;
converting the data includes decrementing the value in the particle storage region for managing the number of particles corresponding to each of the plurality of data by the number of particles set to the storage regions corresponding to the vertexes of the graph; and
initializing the number of particles set to the storage region corresponding to the deletion vertex after the updating process is executed; and
incrementing the value in the particle storage region, which manages the number of particles set to the storage region corresponding to the deletion vertex, by the number of particles set to the storage region corresponding to the deletion vertex.

5. The data generation method according to claim 4 , wherein
at least one of the computers is programmed to identify a class to which the identification target belongs using the pattern data.

6. A computer system, comprising:
a plurality of computers each including a processor, a memory connected to the processor, and a network interface connected to the processor;
at least one of the computers including a processor programmed to acquire a plurality of time series data and generate pattern data representative of a feature value for identifying a class to which an identification target belongs using the plurality of time series data;
at least one of the computers including a memory configured to retain graph information for managing a graph configured from a plurality of vertexes and sides which connect the plurality of vertexes to each other;
the processor is programmed to
acquire the plurality of data and the graph information and assure storage regions in number equal to the number of vertexes included in the graph for storing the plurality of data,
convert each of the plurality of time series data into an input value and set at least one input value to a storage region corresponding to at least one of the vertexes included in the graph,
execute an updating process for updating a value set to a storage region corresponding to a first vertex using the value set to the storage region corresponding to the first vertex and a value set to a storage region corresponding to a different vertex directly connected to the first vertex, and
output a set of values set to the storage regions individually corresponding to the plurality of vertexes included in the graph as the pattern data, the pattern data being the feature value generated from the plurality of time series data and being given as the set of values set to the vertexes of the graph, the pattern data being indicative of a transient response and being the feature value on which an influence between the plurality of time series data and a transient of the time series data are reflected.

7. The computer system according to claim 6 , wherein
the memory retains definition information of the vertexes to which the plurality of input values are individually set; and
the processor is programmed to
specify, when the input values are to be set to the storage regions corresponding to the vertexes, a vertex to which each of the plurality of input values is to be set based on the definition information,
update, in the updating process, the value set to the storage region corresponding to the first vertex based on a reaction diffusion equation in which the value set to the storage region corresponding to the first vertex and the value set to the storage region corresponding to the different vertex connected to the first vertex are used, and
output a set of values set to the storage regions indivi

Assignees

• Hitachi Ltd

Inventors

• Miyakoshi Junichi
• Yamaoka Masanao
• Takahashi Hiromasa
• Ho Shirun
• Kurotsuchi Kenzo
• NAGATA Sanato

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06F16/9024 Primary

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06F16/2379

  Updates performed during online database operations; commit processing · CPC title