Image generation method based on adaptive gradient cutting under differential privacy protection

A differential privacy, image generation technology, applied in 2D image generation, image enhancement, image analysis and other directions, can solve problems such as unreasonable cropping and adding noise, and achieve the effect of fast training convergence and high image quality.

Pending Publication Date: 2022-03-11
BEIHANG UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0049] The invention discloses an image generation method based on adaptive gradient cropping under differential privacy protection, which solves the problem of excessive noise and unreasonable cropping in the previous generation model algorithm based on privacy protection under the gradient clipping method based on training data. At the same time, the distributio

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
  • Image generation method based on adaptive gradient cutting under differential privacy protection
  • Image generation method based on adaptive gradient cutting under differential privacy protection
  • Image generation method based on adaptive gradient cutting under differential privacy protection

Examples

Experimental program
Comparison scheme
Effect test

Example Embodiment

[0061] Generate huge amounts of data subject to network trained to generate photorealistic image data, and the training process gradient or after training network weights are leaked reasoning model could allow an attacker to attack members of reasoning or an attack by guessing whether the training data data. Therefore, to ensure the privacy of the training process to generate a model, generating a network comprising a two-part structure, respectively, is determined and generator. Both the neural network, thus training process is similar to the training data is approximately the input network obtained results, the calculated loss function based on the results obtained derivative of the loss function or the gradient discriminator generator. According to the nature of the post-processing differential privacy, just training process satisfies the differential discriminator privacy protection, generator training process naturally also meet the definition of differential privacy. So befo...

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 an image generation method based on adaptive gradient cutting under differential privacy protection, and solves the problems of excessive noise and unreasonable cutting in a gradient cutting method based on training data in a conventional privacy protection-based model generation algorithm. Meanwhile, a cutting threshold distribution method based on a model layer gradient cutting method is improved, the efficiency of network learning is improved on the premise that the cutting threshold is reasonably distributed, and the magnitude of gradient noise added after cutting is further reduced under the condition that finer cutting granularity is selected. Therefore, the method can provide a more reasonable gradient cutting mode and lower noise addition. Compared with the prior art, the method has the advantages that the generation model can be more quickly converged to a local optimal solution under the conditions of more appropriate cutting and smaller noise addition, and the purpose of generating a high-quality picture is achieved.

Description

technical field [0001] The present invention relates to the field of differential privacy and artificial intelligence. It uses differential privacy technology to ensure the privacy of training data in machine learning scenarios, and uses adaptive tailoring technology to ensure the availability of the adversarial generation network model after gradient noise. A privacy-preserving approach to high-quality image generation. Background technique [0002] Machine learning is an integral part of today's society. Every aspect of people's lives, from food delivery to short videos, to medical education, cannot do without the help of machine learning algorithms. Although machine learning facilitates human society, this convenience is not without cost. Machine learning models require a large amount of training data to ensure the availability and robustness of the model. However, the current means of attacking machine learning can already infer whether a certain data is in the trainin...

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
IPC IPC(8): G06T11/00G06N3/04G06N3/08
CPCG06T11/00G06N3/08G06T2207/20132G06T2207/20004G06N3/045
Inventor 姚燕青林建辰翟征德
Owner BEIHANG UNIV
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