Differentially private GANs by adding noise to Discriminator’s loss

Han, Chunling; Xue, Rui

doi:10.1016/j.cose.2021.102322

Cited by 12 publications

(6 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is based on simultaneously training models G and D. G is a generative model that tries to catch the data distribution. D is a discriminative model that simultaneously attempts to guess the probability that a sample came from the training data rather than G. Some research uses GAN for privacy protection by creating a synthetic copy of the original data [34], [35], [36], and [37].…”

Section: B Gan Implementations With Iot Data Protection Approachmentioning

confidence: 99%

A Hybrid Approach With GAN and DP for Privacy Preservation of IIoT Data

Hindistan

Yetkin

2023

IEEE Access

View full text Add to dashboard Cite

There are emerging trends to use the Industrial Internet of Things (IIoT) in manufacturing and related industries. Machine Learning (ML) techniques are widely used to interpret the collected IoT data for improving the company's operational excellence and predictive maintenance. In general, ML applications require high computational resource allocation and expertise. Manufacturing companies usually transfer their IIoT data to an ML-enabled third party or a cloud system. Although the transmission process uses encryption, ML applications still need decrypted data to perform ML tasks efficiently. Therefore, the third parties may have unacceptable access rights during the data processing to the content of IIoT data that contains a portrait of the production process. IIoT data may include hidden sensitive features, creating an information leakage for the companies. All these concerns prevent companies from sharing their IIoT data with third parties. This paper proposes a novel method based on the hybrid usage of Generative Adversarial Networks (GAN) and Differential Privacy (DP) to preserve sensitive data in IIoT operations. We aim to sustain IIoT data privacy with minimal accuracy loss without adding high additional computational costs to the overall data processing scheme. We demonstrate the efficiency of our approach with publicly available data sets and a realistic IIoT data set collected from a confectionery production process.

show abstract

Section: B Gan Implementations With Iot Data Protection Approachmentioning

confidence: 99%

A Hybrid Approach With GAN and DP for Privacy Preservation of IIoT Data

Hindistan

Yetkin

2023

IEEE Access

View full text Add to dashboard Cite

show abstract

“…It is also necessary to pay attention to the effectiveness of adding noise when perturbing the objective function. For example, differentially private GAN [33] adds noise directly to the discriminator loss, and the noise factor as a constant term in the reverse derivation has no variations. Thus, the privacy of each parameter cannot be guaranteed.…”

Section: Related Workmentioning

confidence: 99%

“…Unlike prior work on adding noise to discriminator gradients, we focus on the interaction between the generator and discriminator because the network is more vulnerable to infection, and the adversary has more accessible information. Adding noise to the cost function is more straightforward and easy [33], and it is easier to operate without directly exchanging the parameters of the discriminator-generator interaction or modify the gradient update process. From [33], we consider that the noise on the cost function cannot be unfunctional during the parameters optimization.…”

Section: Figure 1 the Overview Of Dpba-wgan Structurementioning

confidence: 99%

“…Adding noise to the cost function is more straightforward and easy [33], and it is easier to operate without directly exchanging the parameters of the discriminator-generator interaction or modify the gradient update process. From [33], we consider that the noise on the cost function cannot be unfunctional during the parameters optimization. Thus, the noise term is added as a common factor of cost so that the noise can always contribute during backpropagation.…”

Section: Figure 1 the Overview Of Dpba-wgan Structurementioning

confidence: 99%

See 1 more Smart Citation

DPBA-WGAN: A Vector-Valued Differential Private Bilateral Alternative Scheme on WGAN for Image Generation

Zhang

2023

IEEE Access

View full text Add to dashboard Cite

The large amount of sensitive personal information used in deep learning models has attracted considerable attention for privacy security. Sensitive data may be memorialized or encoded into the parameters or generation of the Wasserstein Generative Adversarial Networks (WGAN), which can be prevented by implementing privacy-preserving algorithms during the parameter training process. Meanwhile, the model is also expected to obtain effective generated results. We propose a vector-valued differential private bilateral alternative (DPBA) algorithm, a novel perturbation method for the training process. The vector-valued Gaussian (VVG) noise involving functional structure information is injected into the WGAN to generate data with privacy protection, and the model is verified to satisfy differential privacy. The bilateral alternative noise can eventually randomly perturb the gradient and generates informative feature-rich samples. The dynamic noise and vector-based perturbation approach ensures privacy strength. After extensive evaluation, our algorithm outperformed state-of-the-art techniques in terms of usability metrics for all validation datasets. The downstream classification accuracy for the generated Modified National Institute of Standards and Technology (Mnist) was 97.04%, whereas that for the Fashion Modified National Institute of Standards and Technology (Fashion-Mnist) dataset was 80.91%. Mnist improved the average accuracy of the neural network classifier by at least 16.81%, and Fashion-Mnist by at least 3.55%. In the multichannel generation tasks, the binary classification accuracy improved by at least 10.4% compared to CelebFaces Attributes (CelebA), and the accuracy of the Street View House Numbers (SVHN) was as high as 86.1%. The perturbation method proved highly resilient to gradient attack recovery under simulated gradient attacks.

show abstract

“…The emergence of Generative Adversarial Networks (GANs) [26] provides a new path for face de-identification research [15,[27][28][29][30]. In [31], a GAN was used for the inpainting of facial landmarks in the conditioned heads.…”

Section: Related Workmentioning

confidence: 99%

A Novel Generative Model for Face Privacy Protection in Video Surveillance with Utility Maintenance

et al. 2022

View full text Add to dashboard Cite

In recent years, the security and privacy issues of face data in video surveillance have become one of the hotspots. How to protect privacy while maintaining the utility of monitored faces is a challenging problem. At present, most of the mainstream methods are suitable for maintaining data utility with respect to pre-defined criteria such as the structure similarity or shape of the face, which bears the criticism of poor versatility and adaptability. This paper proposes a novel generative framework called Quality Maintenance-Variational AutoEncoder (QM-VAE), which takes full advantage of existing privacy protection technologies. We innovatively add the loss of service quality to the loss function to ensure the generation of de-identified face images with guided quality preservation. The proposed model automatically adjusts the generated image according to the different service quality evaluators, so it is generic and efficient in different service scenarios, even some that have nothing to do with simple visual effects. We take facial expression recognition as an example to present experiments on the dataset CelebA to demonstrate the utility-preservation capabilities of QM-VAE. The experimental data show that QM-VAE has the highest quality retention rate of 86%. Compared with the existing method, QM-VAE generates de-identified face images with significantly improved utility and increases the effect by 6.7%.

show abstract

Differentially private GANs by adding noise to Discriminator’s loss

Cited by 12 publications

References 2 publications

A Hybrid Approach With GAN and DP for Privacy Preservation of IIoT Data

A Hybrid Approach With GAN and DP for Privacy Preservation of IIoT Data

DPBA-WGAN: A Vector-Valued Differential Private Bilateral Alternative Scheme on WGAN for Image Generation

A Novel Generative Model for Face Privacy Protection in Video Surveillance with Utility Maintenance

Contact Info

Product

Resources

About