Method for a generative adversarial network with different hierarchical function combinations

A network and function technology, applied in the field of confrontation generation network with different levels of function combinations, can solve the problems of accuracy, generation efficiency, and no comprehensive consideration of features and gradients, so as to improve the poisoning effect, enhance the poisoning ability, and ensure the attack effect of effect

Pending Publication Date: 2022-01-28
BEIJING UNIV OF POSTS & TELECOMM
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

For example, DeepConfuse can successfully deceive the image recognition system. However, since the influence of features, gradients and other levels are not comprehensively considered, the accuracy and generation efficiency still need to be improved.

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
  • Method for a generative adversarial network with different hierarchical function combinations
  • Method for a generative adversarial network with different hierarchical function combinations
  • Method for a generative adversarial network with different hierarchical function combinations

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0060] Since deep learning-based image recognition systems rely on the training of neural networks, which are vulnerable to adversarial attacks and cause misjudgment, the security of deep learning systems is vulnerable to carefully crafted adversarial examples. Researchers continue to propose new methods of adversarial attacks, but while neural networks are powerful enough to learn powerful models in the presence of natural noise, collecting data from untrusted sources makes neural networks vulnerable to successful attacks. The present invention proposes an adversarial generative network (DPGAN) method using different levels of function combinations, which destroys training data through data poisoning and ultimately destroys the entire learning process, thereby effectively improving the poisoning effect of poisoned samples.

[0061] In order to better understand the technical solution, the method of the present invention will be described in detail below in conjunction with the...

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

The invention discloses a method for a generative adversarial network with different hierarchical function combinations, training data is destroyed through data poisoning, and finally the whole learning process is destroyed, so that the poisoning effect of a poisoning sample is effectively improved. According to the method, a combined loss function is used, and data are processed in different stages, so that the purpose of misrecognition is achieved. Meanwhile, in order to further achieve the effect of picture misrecognition, comparative learning in the field of self-supervised learning is applied to the process of intelligent model training. According to the method, a data pair mode is used during data loading, and a feature integration mode is used in a training process to process the data, so that the poison ability during data pollution is enhanced, and the attack effect under the condition that an attack pattern is not identified by naked eyes as far as possible is ensured.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for confronting generation networks with different level function combinations. Background technique [0002] In recent years, with the advancement of deep learning technology, it has even surpassed human-level performance in many tasks, so it has been widely used in daily life, such as face recognition, fingerprint recognition and network security, as well as those with high failure costs. applications such as autonomous driving. However, studies have shown that deep neural networks are extremely vulnerable to adversarial examples. By adding some small perturbations to images or detection targets, attackers can cause neural networks to learn inaccurate features and eventually get wrong prediction results. The process is called an "adversarial attack," in which an adversarial example is a modified version of a clean image that is deliberately perturbed (such as...

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): G06N3/08G06N3/04G06K9/62
CPCG06N3/084G06N3/04G06F18/241
Inventor 刘亮郑霄龙刘知瑶马华东
Owner BEIJING UNIV OF POSTS & TELECOMM
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap