Unlock instant, AI-driven research and patent intelligence for your innovation.

Evaluation device, evaluation method, and evaluation program

An evaluation device and evaluation method technology, applied in neural learning methods, computer security devices, biological neural network models, etc.

Pending Publication Date: 2021-10-22
NIPPON TELEGRAPH & TELEPHONE CORP
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] However, in the past, the evaluation of Adversarial Attack's tolerance to VAE could only be performed when labels were attached to the input data of VAE.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Evaluation device, evaluation method, and evaluation program
  • Evaluation device, evaluation method, and evaluation program
  • Evaluation device, evaluation method, and evaluation program

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0023] Hereinafter, modes (embodiments) for carrying out the present invention will be described with reference to the drawings. The present invention is not limited to the embodiments described below.

[0024] First, use figure 1 The outline of VAE will be described. Such as figure 1 As shown, VAE converts the input data into low-dimensional hidden variables through a neural network called an encoder, and then learns through a decoder to reconstruct the input data. As a result of the above learning, the information required to reconstruct the input data is stored in the latent variables of the VAE. That is, latent variables exhibit essential characteristics of the input data.

[0025] Thus, the learned VAE, for example, accepts figure 2 When the data group shown by symbol 201 is input, the data group is reconstructed, and the data group shown by symbol 202 is output. However, if noise is added to VAE by Adversarial Attack, VAE may not be able to reconstruct the data gr...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

An evaluation device of the present invention receives input of latent variables of a variational auto-encoder, clusters the latent variables that were input, and for each cluster, assigns a label indicating the cluster to the latent variables belonging to the cluster. Afterwards, the evaluation device implements learning of a classifier such that the latent variables are accurately classified on the basis of the assigned label, subjects the classifier after learning to an evaluation of resistance to an adversarial attack, and outputs the results of the resistance evaluation. Through this, the evaluation device can evaluate the resistance to an adversarial attack even with a variational auto-encoder that uses, as input data, data without a label.

Description

technical field [0001] The present invention relates to an evaluation device, an evaluation method, and an evaluation program. Background technique [0002] As a method of deep learning, VAE (Variational Auto Encoder, variational autoencoder) is widely used, but in recent years, the threat of Adversarial Attack (adversarial attack), which is an attack against the vulnerability of machine learning, has been hinted. Adversarial Attack refers to an attack method that degrades the quality of learning and misclassifies classification problems by consciously giving machine learning a vulnerability. In recent years, in the teaching classification problem based on deep learning, the method of using the Adversarial Example (adversarial example) containing the minute noise of the data that cannot recognize the person to make the misrecognition of the data a threat is being researched. [0003] Adversarial Attack using Adversarial Example is also a great threat to VAE, which is learni...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06N3/02
CPCG06N3/088G06N20/00G06N3/047G06N3/045G06F21/552G06F2221/034
Inventor 高桥知克山田真德
Owner NIPPON TELEGRAPH & TELEPHONE CORP