Synthetic Data -- what, why and how?
Giovanni Cherubin2022, arXiv (Cornell University)
Sign up for access to the world's latest research
checkGet notified about relevant papers
checkSave papers to use in your research
checkJoin the discussion with peers
checkTrack your impact
Abstract
AI
AI
Synthetic data technologies are gaining traction due to their applications in privacy, fairness, and data augmentation. Despite their promise, they come with inherent risks, particularly regarding privacy guarantees and the representation of outliers. Effective deployment requires careful consideration of these factors to ensure that synthetic data serves as a valuable tool in research without compromising individual privacy.
Related papers
Enabling Synthetic Data adoption in regulated domains
2022
The switch from a Model-Centric to a Data-Centric mindset is putting emphasis on data and its quality rather than algorithms, bringing forward new challenges. In particular, the sensitive nature of the information in highly regulated scenarios needs to be accounted for. Specific approaches to address the privacy issue have been developed, as Privacy Enhancing Technologies. However, they frequently cause loss of information, putting forward a crucial trade-off among data quality and privacy. A clever way to bypass such a conundrum relies on Synthetic Data: data obtained from a generative process, learning the real data properties. Both Academia and Industry realized the importance of evaluating synthetic data quality: without all-round reliable metrics, the innovative data generation task has no proper objective function to maximize. Despite that, the topic remains under-explored. For this reason, we systematically catalog the important traits of synthetic data quality and privacy, and devise a specific methodology to test them. The result is DAISYnt (aDoption of Artificial Intelligence SYnthesis): a comprehensive suite of advanced tests, which sets a de facto standard for synthetic data evaluation. As a practical use-case, a variety of generative algorithms have been trained on real-world Credit Bureau Data. The best model has been assessed, using DAISYnt on the different synthetic replicas. Further potential uses, among others, entail auditing and fine-tuning of generative models or ensuring high quality of a given synthetic dataset. From a prescriptive viewpoint, eventually, DAISYnt may pave the way to synthetic data adoption in highly regulated domains, ranging from Finance to Healthcare, through Insurance and Education.
Synthetic Data: Can We Trust Statistical Estimators?
arXiv (Cornell University), 2023
Recent advances in generative models facilitate the creation of synthetic data to be made available for research in privacy-sensitive contexts. However, the analysis of synthetic data raises a unique set of methodological challenges. In this work, we highlight the importance of inferential utility and provide empirical evidence against naive inference from synthetic data, whereby synthetic data are treated as if they were actually observed. Before publishing synthetic data, it is essential to develop statistical inference tools for such data. By means of a simulation study, we show that the rate of falsepositive findings (type 1 error) will be unacceptably high, even when the estimates are unbiased. Despite the use of a previously proposed correction factor, this problem persists for deep generative models, in part due to slower convergence of estimators and resulting underestimation of the true standard error. We further demonstrate our findings through a case study.
Use of Synthetic Data to Train AI Models
Policy Guideline produced by the United Nations University, 2024
Using synthetic or artiffcially generated data in training Artiffcial Intelligence (AI) algorithms is a burgeoning practice with signiffcant potential to affect society directly. It can address data scarcity, privacy, and bias issues but does raise concerns about data quality, security, and ethical implications. While some systems use only synthetic data, most times synthetic data is used together with real-world data to train AI models. Our recommendations in this document are for any system where some synthetic data are used. The use of synthetic data has the potential to enhance existing data to allow for more efffcient and inclusive practices and policies. However, we cannot assume synthetic data to be automatically better or even equivalent to data from the physical world. There are many risks to using synthetic data, including cybersecurity risks, bias propagation, and increasing model error. This document sets out recommendations for the responsible use of synthetic data in AI training.
A Multi-Dimensional Evaluation of Synthetic Data Generators
IEEE Access, 2022
Synthetic datasets are gradually emerging as solutions for data sharing. Multiple synthetic data generators have been introduced in the last decade fueled by advancement in machine learning and by the increased demand for fast and inclusive data sharing, yet their utility is not well understood. Prior research tried to compare the utility of synthetic data generators using different evaluation metrics. These metrics have been found to generate conflicting conclusions making direct comparison of synthetic data generators very difficult. This paper identifies four criteria (or dimensions) for masked data evaluation by classifying available utility metrics into different categories based on the measure they attempt to preserve: attribute fidelity, bivariate fidelity, population fidelity, and application fidelity. A representative metric from each category is chosen based on popularity and consistency, and the four metrics are used to compare the overall utility of four recent data synthesizers across 19 datasets of different sizes and feature counts. The paper also examines correlations between the selected metrics in an attempt to streamline synthetic data utility.
Synthetic Data - A Privacy Mirage
ArXiv, 2020
Synthetic datasets produced by generative models are advertised as a silver-bullet solution to privacy-preserving data sharing. Claims about the privacy benefits of synthetic data, however, have not been supported by a rigorous privacy analysis. In this paper, we introduce an evaluation framework that enables data holders to (I) quantify the privacy gain of publishing a synthetic dataset instead of the raw data, and (II) compare the privacy properties of generative model training algorithms. We illustrate the utility of the framework and quantify privacy gain with respect to two concerns, the risk of re-identification via linkage and the risk of attribute disclosure, on synthetic data produced by a range of generative models, from simple independent histograms to differentially private GANs. We find that, across the board, synthetic data provides little privacy gain even under a black-box adversary with access to a single synthetic dataset only. Moreover, we observe that some target...
Measuring the quality of Synthetic data for use in competitions
ArXiv, 2018
Machine learning has the potential to assist many communities in using the large datasets that are becoming more and more available. Unfortunately, much of that potential is not being realized because it would require sharing data in a way that compromises privacy. In order to overcome this hurdle, several methods have been proposed that generate synthetic data while preserving the privacy of the real data. In this paper we consider a key characteristic that synthetic data should have in order to be useful for machine learning researchers - the relative performance of two algorithms (trained and tested) on the synthetic dataset should be the same as their relative performance (when trained and tested) on the original dataset.
Generation and evaluation of privacy preserving synthetic health data
Neurocomputing, 2020
We develop metrics for measuring the quality of synthetic health data for both education and research. We use novel and existing metrics to capture a synthetic dataset's resemblance, privacy, utility and footprint. Using these metrics, we develop an end-to-end workflow based on our generative adversarial network (GAN) method, HealthGAN, that creates privacy preserving synthetic health data. Our workflow meets privacy specifications of our data partner: (1) the HealthGAN is trained inside a secure environment; (2) the HealthGAN model is used outside of the secure environment by external users to generate synthetic data. This second step facilitates data handling for external users by avoiding de-identification, which may require special user training, be costly, or cause loss of data fidelity. This workflow is compared against five other baseline methods. While maintaining resemblance and utility comparable to other methods, HealthGAN provides the best privacy and footprint. We present two case studies in which our methodology was put to work in the classroom and research settings. We evaluate utility in the classroom through a data analysis challenge given to students and in research by replicating three different medical papers with synthetic data. Data, code, and the challenge that we organized for educational purposes are available.
Multiply-Imputed Synthetic Data: Advice to the Imputer
Journal of Official Statistics, 2017
Several statistical agencies have started to use multiply-imputed synthetic microdata to create public-use data in major surveys. The purpose of doing this is to protect the confidentiality of respondents’ identities and sensitive attributes, while allowing standard complete-data analyses of microdata. A key challenge, faced by advocates of synthetic data, is demonstrating that valid statistical inferences can be obtained from such synthetic data for non-confidential questions. Large discrepancies between observed-data and synthetic-data analytic results for such questions may arise because of uncongeniality; that is, differences in the types of inputs available to the imputer, who has access to the actual data, and to the analyst, who has access only to the synthetic data. Here, we discuss a simple, but possibly canonical, example of uncongeniality when using multiple imputation to create synthetic data, which specifically addresses the choices made by the imputer. An initial, unan...
Assessing privacy and quality of synthetic health data
Proceedings of the Conference on Artificial Intelligence for Data Discovery and Reuse, 2019
This paper builds on the results of the ESANN 2019 conference paper "Privacy Preserving Synthetic Health Data" [16], which develops metrics for assessing privacy and utility of synthetic data and models. The metrics laid out in the initial paper show that utility can still be achieved in synthetic data while maintaining both privacy of the model and the data being generated. Specifically, we focused on the success of the Wasserstein GAN method, renamed HealthGAN, in comparison to other data generating methods. In this paper, we provide additional novel metrics to quantify the susceptibility of these generative models to membership inference attacks [14]. We also introduce Discriminator Testing, a new method of determining whether the different generators overfit on the training data, potentially resulting in privacy losses. These privacy issues are of high importance as we prepare a final workflow for generating synthetic data based on real data in a secure environment. The results of these tests complement the initial tests as they show that the Parzen windows method, while having a low privacy loss in adversarial accuracy metrics, fails to preserve privacy in the membership inference attack. Only HealthGAN shows both an optimal value for privacy loss and the membership inference attack. The discriminator testing adds to the confidence as HealthGAN retains resemblance to the training data, without reproducing the training data.
On the Generation of Privatized Synthetic Data Using Distance Transforms
Organizations have interest in research collaboration efforts that involve data sharing with peers. However, such partnerships often come with confidentiality risks that could involve insider attacks and untrustworthy collaborators who might leak sensitive information. To mitigate such data sharing vulnerabilities, entities share privatized data with retracted sensitive information. However, while such data sets might offer some assurances of privacy, maintaining the statistical traits of the original data, is often problematic, leading to poor data usability. Therefore, in this paper, a confidential synthetic data generation heuristic, that employs a combination of data privacy and distance transforms techniques, is presented. The heuristic is used for the generation of privatized numeric synthetic data, while preserving the statistical traits of the original data. Empirical results from applying unsupervised learning, using k-means, to test the usability of the privatized synthetic data set, are presented. Preliminary results from this implementation show that it might be possible to generate privatized synthetic data sets, with the same statistical morphological structure as the original, using data privacy and distance transforms methods.
Related topics
Related papers
Synthetic Data -- Anonymisation Groundhog Day
2020
1 Synthetic data has been advertised as a silver-bullet solution to privacy-preserving data publishing that addresses the shortcomings of traditional anonymisation techniques. The promise is that synthetic data drawn from generative models preserves the statistical properties of the original dataset but, at the same time, provides perfect protection against privacy attacks. In this work, we present the first quantitative evaluation of the privacy gain of synthetic data publishing and compare it to that of previous anonymisation techniques. Our evaluation of a wide range of state-of-the-art generative models demonstrates that synthetic data either does not prevent inference attacks or does not retain data utility. In other words, we empirically show that synthetic data does not provide a better tradeoff between privacy and utility than traditional anonymisation techniques. Furthermore, in contrast to traditional anonymisation, the privacy-utility tradeoff of synthetic data publishing...
Privacy Measurement in Tabular Synthetic Data: State of the Art and Future Research Directions
arXiv (Cornell University), 2023
Synthetic data (SD) have garnered attention as a privacy enhancing technology. Unfortunately, there is no standard for quantifying their degree of privacy protection. In this paper, we discuss proposed quantification approaches. This contributes to the development of SD privacy standards; stimulates multi-disciplinary discussion; and helps SD researchers make informed modeling and evaluation decisions. To the best of our knowledge, there is no widely accepted definition of SD. Following Jordon et al. [4], we propose Definition 2.1. Definition 2.1. (Synthetic data,) Synthetic data (SD) is data that has been generated using a purpose-built mathematical model or algorithm (the "generator"), with the aim of solving a (set of) data science task(s).
Assessing the utility of synthetic data: A density ratio perspective
Zenodo (CERN European Organization for Nuclear Research), 2023
Synthetic data can be a solution to reduce disclosure risks that arise when disseminating research data to the public. However, for the synthetic data to be useful for general inferential purposes, it is paramount that its distribution is similar to the distribution of the observed data. Often, data disseminators consider multiple synthetic data models and make refinements in an iterative fashion. After each adjustment, it is crucial to evaluate whether the quality of the synthetic data has actually improved. Although many methods exist to provide such an evaluation, their results are often incomplete or even misleading. To improve the evaluation strategy for synthetic data, and thereby the quality of synthetic data itself, we propose to use the density ratio estimation framework. Using techniques from this field, we show how an interpretable utility measure can be obtained from the ratio of the observed and synthetic data densities. We show how the density ratio estimation framework bridges the gap between fit-for-purpose and global utility measures, and discuss how it can also be used to evaluate analysis-specific utility. Using empirical examples, we show that density ratio estimation improves on existing (global) utility measures by providing higher statistical power and offering a fine-grained view of discrepancies between the observed and synthetic data. Moreover, we describe several additional advantages of the approach, such as providing a measure of utility on the level of individual synthetic data points, automatic model selection without requiring user specification, and readily available high-dimensional extensions. We conclude that density ratio estimation provides a promising framework in synthetic data generation workflows and present an R-package with functionality to implement the approach.
Representative & Fair Synthetic Data
2021
Algorithms learn rules and associations based on the training data that they are exposed to. Yet, the very same data that teaches machines to understand and predict the world, contains societal and historic biases, resulting in biased algorithms with the risk of further amplifying these once put into use for decision support. Synthetic data, on the other hand, emerges with the promise to provide an unlimited amount of representative, realistic training samples, that can be shared further without disclosing the privacy of individual subjects. We present a framework to incorporate fairness constraints into the self-supervised learning process, that allows to then simulate an unlimited amount of representative as well as fair synthetic data. This framework provides a handle to govern and control for privacy as well as for bias within AI at its very source: the training data. We demonstrate the proposed approach by amending an existing generative model architecture and generating a repr...
Comparing the Utility and Disclosure Risk of Synthetic Data with Samples of Microdata
Lecture Notes in Computer Science, 2022
Most statistical agencies release randomly selected samples of Census microdata, usually with sample fractions under 10% and with other forms of statistical disclosure control (SDC) applied. An alternative to SDC is data synthesis, which has been attracting growing interest, yet there is no clear consensus on how to measure the associated utility and disclosure risk of the data. The ability to produce synthetic Census microdata, where the utility and associated risks are clearly understood, could mean that more timely and wider-ranging access to microdata would be possible. This paper follows on from previous work by the authors which mapped synthetic Census data on a risk-utility (R-U) map. The paper presents a framework to measure the utility and disclosure risk of synthetic data by comparing it to samples of the original data of varying sample fractions, thereby identifying the sample fraction which has equivalent utility and risk to the synthetic data. Three commonly used data synthesis packages are compared with some interesting results. Further work is needed in several directions but the methodology looks very promising.
Bias on Demand: A Modelling Framework That Generates Synthetic Data With Bias
2023 ACM Conference on Fairness, Accountability, and Transparency
Nowadays, Machine Learning (ML) systems are widely used in various businesses and are increasingly being adopted to make decisions that can significantly impact people's lives. However, these decision-making systems rely on data-driven learning, which poses a risk of propagating the bias embedded in the data. Despite various * The views and opinions expressed are those of the authors and do not necessarily reflect the views of Intesa Sanpaolo, its affiliates or its employees. This work is licensed under a Creative Commons Attribution International 4.0 License.
A Study of Privatized Synthetic Data Generation Using Discrete Cosine Transforms
International Journal of Advanced Computer Science and Applications, 2014
In order to comply with data confidentiality requirements, while meeting usability needs for researchers, entities are faced with the challenge of how to publish privatized data sets that preserve the statistical traits of the original data. One solution to this problem, is the generation of privatized synthetic data sets. However, during data privatization process, the usefulness of data, have a propensity to diminish even as privacy might be guaranteed. Furthermore, researchers have documented that finding an equilibrium between privacy and utility is intractable, often requiring trade-offs. Therefore, as a contribution, the Filtered Classification Error Gauge heuristic, is presented. The suggested heuristic is a data privacy and usability model that employs data privacy, signal processing, and machine learning techniques to generate privatized synthetic data sets with acceptable levels of usability. Preliminary results from this study show that it might be possible to generate privacy compliant synthetic data sets using a combination of data privacy, signal processing, and machine learning techniques, while preserving acceptable levels of data usability.
dsSynthetic: Synthetic data generation for the DataSHIELD federated analysis system
2022
Objective. Platforms such as DataSHIELD allow users to analyse sensitive data remotely, without havingfull access to the detailed data items (federated analysis). While this feature helps to overcome difficultieswith data sharing, it can make it challenging to write code without full visibility of the data. One solutionis to generate realistic, non-disclosive synthetic data that can be transferred to the analyst so they canperfect their code without the access limitation. When this process is complete, they can run the code on the real data. Results. We have created a package in DataSHIELD (dsSynthetic) which allows generation of realistic synthetic data, building on existing packages. In our paper and accompanying tutorial we demonstratehow the use of synthetic data generated with our package can help DataSHIELD users with tasks suchas writing analysis scripts and harmonising data to common scales and measures.
Synthetic Data Generation using Benerator Tool
Datasets of different characteristics are needed by the research community for experimental purposes. However, real data may be difficult to obtain due to privacy concerns. Moreover, real data may not meet specific characteristics which are needed to verify new approaches under certain conditions. Given these limitations, the use of synthetic data is a viable alternative to complement the real data. In this report, we describe the process followed to generate synthetic data using Benerator, a publicly available tool. The results show that the synthetic data preserves a high level of accuracy compared to the original data. The generated datasets correspond to microdata containing records with social, economic and demographic data which mimics the distribution of aggregated statistics from the 2011 Irish Census.
A Methodology for Controlling Bias and Fairness in Synthetic Data Generation
Applied sciences, 2022
This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY
Loading Preview
Sorry, preview is currently unavailable. You can download the paper by clicking the button above.