Generation of expanded training data contributing to machine learning for relationship data

Patent metadata

Field	Value
Publication number	US-2020257974-A1
Application number	US-201916728314-A
Country	US
Kind code	A1
Filing date	Dec 27, 2019
Priority date	Jan 11, 2019
Publication date	Aug 13, 2020
Grant date	—

Abstract

An apparatus identifies partial tensor data that contributes to machine learning using tensor data in a tensor format obtained by transforming training data having a graph structure. Based on the partial tensor data and the training data, the apparatus generates expanded training data to be used in the machine learning by expanding the training data.

First claim

What is claimed is:

1 . A non-transitory, computer-readable recording medium having stored therein a data expansion program for causing a computer to execute a process comprising:
identifying partial tensor data that contributes to machine learning using first tensor data in a tensor format obtained by transforming first training data having a graph structure; and based on the partial tensor data and the first training data, generating expanded training data to be used in the machine learning by expanding the first training data.

2 . The non-transitory, computer-readable recording medium of claim 1 , wherein:
the machine learning is learning for inputting tensor data using a neural network; and the identifying includes identifying the partial tensor data from a first element matrix for each of dimensions, the first element matrix being used for generating a core tensor from training data and being optimized during the machine learning.

3 . The non-transitory, computer-readable recording medium of claim 2 , the process further comprising:
learning a linear model that locally approximates an output result of the neural network by using the core tensor, wherein the identifying includes: calculating inner products of respective element matrices of the dimensions and respective regression coefficients of the dimensions, each of the regression coefficients being obtained from the linear model, and identifying a first element for each of the dimensions as the partial tensor data, the first element corresponding to an inner product having a largest value.

4 . The non-transitory, computer-readable recording medium of claim 3 , wherein
the generating includes: generating a second element matrix for each of the dimensions, the second element matrix being obtained by adding the first element to each of the first element matrixes, and generating second tensor data corresponding to the expanded training data by inverse transformation that uses the second element matrix for each of the dimensions and the core tensor extracted from second training data that is an expansion reference.

5 . The non-transitory, computer-readable recording medium of claim 4 , wherein
the generating includes selecting, as the second training data, training data in which a classification probability that is an output result of the neural network is less than a threshold and for which a label for expansion is set.

6 . The non-transitory, computer-readable recording medium of claim 2 , the process further comprising:
executing re-learning of the learned neural network by using the generated expanded training data.

7 . A method performed by a computer, the method comprising:
identifying partial tensor data that contributes to machine learning using tensor data in a tensor format obtained by transforming training data having a graph structure; and based on the partial tensor data and the training data, generating expanded training data to be used in the machine learning by expanding the training data.

8 . An apparatus comprising:
a memory; and a processor coupled to the memory and configured to:
identify partial tensor data that contributes to machine learning using tensor data in a tensor format obtained by transforming training data having a graph structure, and
based on the partial tensor data and the training data, generate expanded training data by expanding the first training data to be used in the machine learning.

Assignees

• Fujitsu Ltd

Inventors

• Yano Shotaro
• Nishino Takuya
• Maruhashi Koji

Classifications

• G06F18/211

  Selection of the most significant subset of features · CPC title

• G06F18/2415

  based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate · CPC title

• G06F18/2431

  Multiple classes · CPC title

• G06F18/214

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

• G06N3/09

  Supervised learning · CPC title