PrivPfC: differentially private data publication for classification

Su, Dong; Cao, Jianyu; Li, Ninghui; Lyu, Min

doi:10.1007/s00778-017-0492-3

Cited by 13 publications

(12 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…DP proved to provide strong privacy protection while allowing datasets analysis. This is a hot research topic where research work can be divided into two categories: 1) The first category is the works producing pre-processing datasets balancing between accuracy and privacy for a specific learning model, such as frequent itemset mining models [71], [96], [97], and classification and clustering models [53], [72], [98], [99].…”

Section: Privacy-preserving Datasetsmentioning

confidence: 99%

Differential Privacy for Deep and Federated Learning: A Survey

2022

View full text Add to dashboard Cite

Users' privacy is vulnerable at all stages of the deep learning process. Sensitive information of users may be disclosed during data collection, or during training, or even after releasing the trained learning model. Differential privacy (DP) is one of the main approaches proven to ensure strong privacy protection in data analysis. DP protects the users' privacy by adding noise to the original dataset or the learning parameters. Thus, an attacker could not retrieve the sensitive information of an individual involved in the training dataset. In this survey paper, we analyze and present the main ideas based on DP to guarantee users' privacy in deep and federated learning. In addition, we illustrate all types of probability distributions that satisfy the DP mechanism, with their properties and use cases. Furthermore, we bridge the gap in the literature by providing a comprehensive overview of the different variants of DP, highlighting their advantages and limitations. Our study reveals the gap between theory and application, accuracy, and robustness of DP. Finally, we provide several open problems and future research directions.

show abstract

Section: Privacy-preserving Datasetsmentioning

confidence: 99%

Differential Privacy for Deep and Federated Learning: A Survey

2022

View full text Add to dashboard Cite

show abstract

“…Differential privacy was designed for protecting a single data record from being speculated via adding an appropriate amount of random noise before the publication. For example, adding sensitivity to the histogram on the data range (the sensitivity in the histogram is the calibrated Laplace noise) is a typical protection before data publication [21][22][23]. As the number of data dimensions grows, the calculation volume of the high-dimensional histogram increases exponentially.…”

Section: Related Workmentioning

confidence: 99%

Local Differential Privacy Protection of High-Dimensional Perceptual Data by the Refined Bayes Network

et al. 2020

Sensors

View full text Add to dashboard Cite

Although the Crowd-Sensing perception system brings great data value to people through the release and analysis of high-dimensional perception data, it causes great hidden danger to the privacy of participants in the meantime. Currently, various privacy protection methods based on differential privacy have been proposed, but most of them cannot simultaneously solve the complex attribute association problem between high-dimensional perception data and the privacy threat problems from untrustworthy servers. To address this problem, we put forward a local privacy protection based on Bayes network for high-dimensional perceptual data in this paper. This mechanism realizes the local data protection of the users at the very beginning, eliminates the possibility of other parties directly accessing the user’s original data, and fundamentally protects the user’s data privacy. During this process, after receiving the data of the user’s local privacy protection, the perception server recognizes the dimensional correlation of the high-dimensional data based on the Bayes network, divides the high-dimensional data attribute set into multiple relatively independent low-dimensional attribute sets, and then sequentially synthesizes the new dataset. It can effectively retain the attribute dimension correlation of the original perception data, and ensure that the synthetic dataset and the original dataset have as similar statistical characteristics as possible. To verify its effectiveness, we conduct a multitude of simulation experiments. Results have shown that the synthetic data of this mechanism under the effective local privacy protection has relatively high data utility.

show abstract

“…In [13], the relationship between probabilistic inference and differential privacy is also studied. Xu et al [6] propose DPPro and Su et al [8] propose PrivPfC, which are high-dimensional data publishing methods based on the projection technology. In DPPro, they first apply random projection to project a d-dimensional vector representation of a user's feature attributes into a lower k-dimensional space, and then adding Gaussian noise to each obtained vector to achieve the purpose of privacy protection.…”

Section: Related Workmentioning

confidence: 99%

“…A few solutions have been proposed for the research of high-dimensional data publishing [2]- [14], mainly including the probability graph model [2], [4], [5], the threshold filtering technique and the projection technique [6], [8], but there are some problems in these methods.…”

Section: Introductionmentioning

confidence: 99%

“…Reference [2] uses the sparse vector technology to find the correlation of attribute pairs, but the sparse vector technology does not satisfy differential privacy [7]. Reference [8] combines projection histogram and chi-square association test to achieve the purpose of high-dimensional data publishing. However, the projection histogram does not take into account the correlation between attributes, resulting in low publishing accuracy of the published results.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

PPDP-PCAO: An Efficient High-Dimensional Data Releasing Method With Differential Privacy Protection

Zhang

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Privacy protection in data publishing is an extremely important issue that has been the focus of extensive research in recent years. However, the existing methods have a host of limitations, especially for high-dimensional data publishing. Aiming at the problem of poor availability of publishing results caused by ''the curse of dimensionality'' in high-dimensional data publishing, we present PPDP-PCAO (Privacy Preserving Data Publishing with Principal Component Analysis Optimization) method, which can better address the problem of the lower utility of release results because of the high noise introduced by the curse of dimensionality. PPDP-PCAO improves the Principal Component Analysis (PCA) algorithm by employing the attribute importance, and reduces the dimension of the data with the improved PCA, which reduces the time and space cost. PPDP-PCAO introduces the evaluation mechanism based on mutual-information into data release, which evaluates the data generated by setting the different quantities of principal components to determine the optimal quantities. PPDP-PCAO considers the existence of multi-sensitive attributes in highdimensional data, while the traditional methods of allocating privacy budgets cannot satisfy the personalized privacy protection. PPDP-PCAO introduces the sensitivity preference, combines the optimal matching theory, and designs the sensitive attribute hierarchical protection strategy. Extensive experimental results on different real datasets demonstrate that PPDP-PCAO not only guarantees the privacy of published dataset, but also significantly improves the accuracy and data utility than other high-dimensional data publishing methods.

show abstract

PrivPfC: differentially private data publication for classification

Cited by 13 publications

References 26 publications

Differential Privacy for Deep and Federated Learning: A Survey

Differential Privacy for Deep and Federated Learning: A Survey

Local Differential Privacy Protection of High-Dimensional Perceptual Data by the Refined Bayes Network

PPDP-PCAO: An Efficient High-Dimensional Data Releasing Method With Differential Privacy Protection

Contact Info

Product

Resources

About