Perturbed Gibbs Samplers for Generating Large-Scale Privacy-Safe Synthetic Health Data

Park, Yubin; Ghosh, Joydeep; Shankar, Mallikarjun

doi:10.1109/ichi.2013.76

Cited by 21 publications

(24 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(2) Privacy risk evaluation for synthetic table T ′ is also an independent research problem. This paper adopts two commonly-used metrics in the existing works [45,39,41], namely hitting rate and distance to the closest record (DCR). Intuitively, the metrics measure the likelihood that the original data records can be re-identified by an attacker.…”

Section: Problem Formalizationmentioning

confidence: 99%

“…The statistical approach aims at modeling a joint multivariate distribution for a dataset and then generating fake data by sampling from the distribution. To effectively capture dependence between variates, existing works utilize copulas [35,46], Bayesian networks [62,63], Gibbs sampling [45] and Fourier decompositions [12]. Synopses-based approaches, such as wavelets and multi-dimensional sketches, build compact data summary for massive data [19,55], which can be then used for estimating joint distribution.…”

Section: Related Work For Data Synthesismentioning

confidence: 99%

“…Evaluation on privacy protection. We adopt the following two metrics, which are widely used in the existing works [45,39,41] for privacy evaluation.…”

Section: Evaluation Frameworkmentioning

confidence: 99%

See 2 more Smart Citations

Relational data synthesis using generative adversarial networks

et al. 2020

View full text Add to dashboard Cite

The proliferation of big data has brought an urgent demand for privacy-preserving data publishing. Traditional solutions to this demand have limitations on effectively balancing the tradeoff between privacy and utility of the released data. Thus, the database community and machine learning community have recently studied a new problem of relational data synthesis using generative adversarial networks (GAN) and proposed various algorithms. However, these algorithms are not compared under the same framework and thus it is hard for practitioners to understand GAN's benefits and limitations. To bridge the gaps, we conduct so far the most comprehensive experimental study that investigates applying GAN to relational data synthesis. We introduce a unified GAN-based framework and define a space of design solutions for each component in the framework, including neural network architectures and training strategies. We conduct extensive experiments to explore the design space and compare with traditional data synthesis approaches. Through extensive experiments, we find that GAN is very promising for relational data synthesis, and provide guidance for selecting appropriate design solutions. We also point out limitations of GAN and identify future research directions.

show abstract

Section: Problem Formalizationmentioning

confidence: 99%

Section: Related Work For Data Synthesismentioning

confidence: 99%

See 1 more Smart Citation

Relational data synthesis using generative adversarial networks

et al. 2020

View full text Add to dashboard Cite

show abstract

“…As they have claimed, their proposed method synthesises artificial records while maintaining the statistical features of the original records to the maximum extent possible. Using different data mining and statistical analysis methods, they have concluded that the synthetic dataset delivers results that largely similar to the original dataset [15].…”

Section: Synthesised Data In Real-world Applicationsmentioning

confidence: 99%

Partially Synthesised Dataset to Improve Prediction Accuracy

Aljaaf¹,

Al‐Jumeily²,

Hussain³

et al. 2016

Intelligent Computing Theories and Application

View full text Add to dashboard Cite

Abstract. The real world data sources, such as statistical agencies, library databanks and research institutes are the major data sources for researchers. Using this type of data involves several advantages including, the improvement of credibility and validity of the experiment and more importantly, it is related to a real world problems and typically unbiased. However, this type of data is most likely unavailable or inaccessible for everyone due to the following reasons. First, privacy and confidentiality concerns, since the data must to be protected on legal and ethical basis. Second, collecting real world data is costly and time consuming. Third, the data may be unavailable, particularly in the newly arises research subjects. Therefore, many studies have attributed the use of fully and/or partially synthesised data instead of real world data due to simplicity of creation, requires a relatively small amount of time and sufficient quantity can be generated to fit the requirements. In this context, this study introduces the use of partially synthesised data to improve the prediction of heart diseases from risk factors. We are proposing the generation of partially synthetic data from agreed principles using rule-based method, in which an extra risk factor will be added to the real-world data. In the conducted experiment, more than 85% of the data was derived from observed values (i.e., real-world data), while the remaining data has been synthetically generated using a rule-based method and in accordance with the World Health Organisation criteria. The analysis revealed an improvement of the variance in the data using the first two principal components of partially synthesised data. A further evaluation has been conducted using five popular supervised machine-learning classifiers. In which, partially synthesised data considerably improves the prediction of heart diseases. Where the majority of classifiers have approximately doubled their predictive performance using an extra risk factor.

show abstract

“…One approach is to generate synthetic individual participant data similar enough to the original trial data that analyses yield the same answers. Park and Ghosh developed an initial approach to managing privacy threats using a perturbed Gibbs sampler, a method that generates synthetic data with a quantifiable privacy risk (6). Goodfellow et al (7) developed a method entitled Generative Adversarial Networks (GANs) using neural networks to generate realistic data from complex distributions.…”

Section: Introductionmentioning

confidence: 99%

Privacy-preserving generative deep neural networks support clinical data sharing

Beaulieu‐Jones

Williams

et al. 2017

Preprint

View full text Add to dashboard Cite

Abstract:Though it is widely recognized that data sharing enables faster scientific progress, the sensible need to protect participant privacy hampers this practice in medicine. We train deep neural networks that generate synthetic participants closely resembling study participants. Using the SPRINT trial as an example, we show that machine-learning models built from simulated participants generalize to the original dataset. We incorporate differential privacy, which offers strong guarantees on the likelihood that a participant could be identified as a member of the trial. Investigators who have compiled a dataset can use our method to provide a freely accessible public version that enables other scientists to perform discovery-oriented analyses. Generated data can be released alongside analytical code to enable fully reproducible workflows, even when privacy is a concern. By addressing data sharing challenges, deep neural networks can facilitate the rigorous and reproducible investigation of clinical datasets. IntroductionSharing individual-level data from clinical studies remains challenging. The status quo often requires scientists to establish a formal collaboration and execute extensive data usage agreements before sharing data. These requirements slow or even prevent data sharing between researchers in all but the closest collaborations.

show abstract

Perturbed Gibbs Samplers for Generating Large-Scale Privacy-Safe Synthetic Health Data

Cited by 21 publications

References 12 publications

Relational data synthesis using generative adversarial networks

Relational data synthesis using generative adversarial networks

Partially Synthesised Dataset to Improve Prediction Accuracy

Privacy-preserving generative deep neural networks support clinical data sharing

Contact Info

Product

Resources

About