Differentially Private Functional Mechanism for Generative Adversarial Networks

Zhang, Xinyue; Ding, Jiahao; Errapotu, Sai Mounika; Huang, Xiaoxia; Li, Ming; Pan, Miao

doi:10.1109/globecom38437.2019.9014134

Cited by 12 publications

(7 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the privacy-preserving data analysis, the standard privacy metric, Differential privacy (DP) [9,11], is proposed to measure the privacy risk of each data sample in the dataset, and has already been adopted in many machine learning domains [4,8,18,20,24]. Basically, under DP framework, privacy protection is guaranteed by limiting the difference of the distribution of the output regardless of the value change of any one sample in the dataset.…”

Section: Differential Privacymentioning

confidence: 99%

Differentially Private Robust ADMM for Distributed Machine Learning

Ding

Zhang

Chen

et al. 2019

2019 IEEE International Conference on Big Data (Big Data)

Self Cite

View full text Add to dashboard Cite

Due to massive amounts of data distributed across multiple locations, distributed machine learning has attracted a lot of research interests. Alternating Direction Method of Multipliers (ADMM) is a powerful method of designing distributed machine learning algorithm, whereby each agent computes over local datasets and exchanges computation results with its neighbor agents in an iterative procedure. There exists significant privacy leakage during this iterative process if the local data is sensitive. In this paper, we propose a differentially private ADMM algorithm (P-ADMM) to provide dynamic zeroconcentrated differential privacy (dynamic zCDP), by inserting Gaussian noise with linearly decaying variance. We prove that P-ADMM has the same convergence rate compared to the nonprivate counterpart, i.e., O(1/K) with K being the number of iterations and linear convergence for general convex and strongly convex problems while providing differentially private guarantee. Moreover, through our experiments performed on real-world datasets, we empirically show that P-ADMM has the best-known performance among the existing differentially private ADMM based algorithms.• We propose a differentially private ADMM algorithm (P-ADMM) by introducing a Gaussian noise with a linearly decaying variance to address the privacy concerns in distributed machine learning over large datasets. • We introduce a new privacy framework to quantify the privacy leakage in a distributed and iterative setting,

show abstract

Section: Differential Privacymentioning

confidence: 99%

Differentially Private Robust ADMM for Distributed Machine Learning

Ding

Zhang

Chen

et al. 2019

2019 IEEE International Conference on Big Data (Big Data)

Self Cite

View full text Add to dashboard Cite

show abstract

“…In objective function perturbation, existing work injects Laplace noise into the coefficients to construct differentially private loss function in GAN training. Zhang et al [ 76 ] proposed a new privacy protection GAN, which perturbs the coefficients of the objective function by injecting Laplace noise into the latent space based on the function mechanism to ensure the differential privacy of the training data, and it is reliable to generate high-quality real synthetic data samples without divulging the sensitive information in the training dataset.…”

Section: Differential Privacy Synthetic Data Generationmentioning

confidence: 99%

GDP vs. LDP: A Survey from the Perspective of Information-Theoretic Channel

Liu

Peng

Tian

et al. 2022

Entropy

View full text Add to dashboard Cite

The existing work has conducted in-depth research and analysis on global differential privacy (GDP) and local differential privacy (LDP) based on information theory. However, the data privacy preserving community does not systematically review and analyze GDP and LDP based on the information-theoretic channel model. To this end, we systematically reviewed GDP and LDP from the perspective of the information-theoretic channel in this survey. First, we presented the privacy threat model under information-theoretic channel. Second, we described and compared the information-theoretic channel models of GDP and LDP. Third, we summarized and analyzed definitions, privacy-utility metrics, properties, and mechanisms of GDP and LDP under their channel models. Finally, we discussed the open problems of GDP and LDP based on different types of information-theoretic channel models according to the above systematic review. Our main contribution provides a systematic survey of channel models, definitions, privacy-utility metrics, properties, and mechanisms for GDP and LDP from the perspective of information-theoretic channel and surveys the differential privacy synthetic data generation application using generative adversarial network and federated learning, respectively. Our work is helpful for systematically understanding the privacy threat model, definitions, privacy-utility metrics, properties, and mechanisms of GDP and LDP from the perspective of information-theoretic channel and promotes in-depth research and analysis of GDP and LDP based on different types of information-theoretic channel models.

show abstract

“…a well-trained generative adversarial network (GAN) [18], captures the underlying distribution of the real data, which means there is nothing stopping it from accidentally producing a doppelganger of a sensitive record (or a close enough sample), and simply sampling such a model could reveal much about both individual records and specific sensitive features of the training dataset [56,57]. There are a number of linkage attacks specifically developed for GANs [35,[58][59][60], as well as some defences proposed for all data types (images, time-series, structured data) [56,[61][62][63].…”

Section: Leakage For Different Tasksmentioning

confidence: 99%

“…The more traditional applications of DP, outlined in [136] are the DP online learning, [148][149][150] and DP empirical risk minimization, [151][152][153][154][155][156][157]. However, the range of learning tasks that DP was applied to has widened and now includes nearly anything from the federated ML setting [158] to differentially private recurrent language models [159], and even differentially private generative adversarial networks [61,63], specific DP-GAN applications for generating timeseries [62,111], and tabular mixed feature datasets [160].…”

Section: Differential Privacy Surveysmentioning

confidence: 99%

Survey: Leakage and Privacy at Inference Time

Jegorova¹,

Kaul²,

Mayor³

et al. 2021

Preprint

View full text Add to dashboard Cite

Leakage of data from publicly available Machine Learning (ML) models is an area of growing significance as commercial and government applications of ML can draw on multiple sources of data, potentially including users' and clients' sensitive data. We provide a comprehensive survey of contemporary advances on several fronts, covering involuntary data leakage which is natural to ML models, potential malevolent leakage which is caused by privacy attacks, and currently available defence mechanisms. We focus on inference-time leakage, as the most likely scenario for publicly available models. We first discuss what leakage is in the context of different data, tasks, and model architectures. We then propose a taxonomy across involuntary and malevolent leakage, available defences, followed by the currently available assessment metrics and applications. We conclude with outstanding challenges and open questions, outlining some promising directions for future research.

show abstract

Differentially Private Functional Mechanism for Generative Adversarial Networks

Cited by 12 publications

References 6 publications

Differentially Private Robust ADMM for Distributed Machine Learning

Differentially Private Robust ADMM for Distributed Machine Learning

GDP vs. LDP: A Survey from the Perspective of Information-Theoretic Channel

Survey: Leakage and Privacy at Inference Time

Contact Info

Product

Resources

About