Neural network learning device, method, and program

Patent metadata

Field	Value
Publication number	US-11580383-B2
Application number	US-201716481536-A
Country	US
Kind code	B2
Filing date	Mar 16, 2017
Priority date	Mar 16, 2017
Publication date	Feb 14, 2023
Grant date	Feb 14, 2023

Abstract

A large amount of training data is typically required to perform deep network leaning, making it difficult to achieve using a few pieces of data. In order to solve this problem, the neural network device according to the present invention is provided with: a feature extraction unit which extracts features from training data using a learning neural network; an adversarial feature generation unit which generates an adversarial feature from the extracted features using the learning neural network; a pattern recognition unit which calculates a neural network recognition result using the training data and the adversarial feature; and a network learning unit which performs neural network learning so that the recognition result approaches a desired output.

First claim

What is claimed is:

1. A neural network learning device, comprising:
a processor; and
a memory storing executable instructions that, when executed by the processor, causes the processor to perform as:
a feature extraction unit configured to extract features from training data using a neural network being currently learned;
an adversarial feature generation unit configured to generate an adversarial feature by adding, to the extracted features, perturbations so that recognition by the neural network being currently learned becomes difficult;
a pattern recognition unit configured to calculate a recognized result of the neural network using the extracted features and the adversarial feature; and
a network learning unit configured to learn the neural network so that the recognized result approaches a desired output.

2. The neural network learning device as claimed in claim 1 , wherein the adversarial feature generation unit is configured to generate the adversarial feature under a constraint which is represented by a linear combination of the training data.

3. A pattern recognition apparatus configured to perform pattern recognition based on a neural network which is learned by using the neural network learning device claimed in claim 1 .

4. A neural network learning method comprising:
extracting features from training data using a neural network being currently learned;
generating an adversarial feature by adding, to the extracted features, perturbations so that recognition by the neural network being currently learned becomes difficult;
calculating a recognized result of the neural network using the extracted features and the adversarial feature; and
learning the neural network so that the recognized result approaches a desired output.

5. The neural network learning method as claimed in claim 4 , wherein the generating generates the adversarial feature under a constraint which is represented by a linear combination of the training data.

6. A non-transitory computer readable recording medium for storing a neural network learning program for causing a computer to execute:
a process for extracting features from training data using a neural network being currently learned;
a process for generating an adversarial feature by adding, to the extracted features, perturbations so that recognition by the neural network being currently learned becomes difficult;
a process for calculating a recognized result of the neural network using the extracted features and the adversarial feature; and
a process for learning the neural network so that the recognized result approaches a desired output.

7. The non-transitory computer readable recording medium as claimed in claim 6 , wherein the process for generating causes the computer to generate the adversarial feature under a constraint which is represented by a linear combination of the training data.

Assignees

• Nec Corp

Inventors

• Ishii Masato

Classifications

• G06N3/094

  Adversarial learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06V10/774

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

• G06F18/241

  relating to the classification model, e.g. parametric or non-parametric approaches · CPC title