Towards practical differential privacy in data analysis: Understanding the effect of epsilon on utility in private ERM

Li, Yuzhe; Liu, Yong; Li, Bo; Wang, Weiping; Liu, Nan

doi:10.1016/j.cose.2023.103147

Cited by 8 publications

(5 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, to guarantee meaningful DP, ϵ is consistently set smaller than 1.0 and δ is always set smaller than 1/n, where n denotes the size of the database [20]. Moreover, smaller privacy budgets decrease the utility of the system, which is in line with the common sense and leads to a tradeoff between the utility and the privacy; corresponding discussions are studied in the field of machine learning [48], data-processing systems [49], open data sharing [50], etc. The tradeoff is also shown by the experimental results given in Section 5.…”

Section: Remarkmentioning

confidence: 99%

A Differentially Private Framework for the Dynamic Heterogeneous Redundant Architecture System in Cyberspace

Kang,

Zhang,

Jiang

et al. 2024

Electronics

View full text Add to dashboard Cite

With the development of information technology, tremendous vulnerabilities and backdoors have evolved, causing inevitable and severe security problems in cyberspace. To fix them, the endogenous safety and security (ESS) theory and one of its practices, the Dynamic Heterogeneous Redundant (DHR) architecture, are proposed. In the DHR architecture, as an instance of the multi-heterogeneous system, a decision module is designed to obtain intermediate results from heterogeneous equivalent functional executors. However, privacy-preserving is not paid attention to in the architecture, which may cause privacy breaches without compromising the ESS theory. In this paper, based on differential privacy (DP), a theoretically rigorous privacy tool, we propose a privacy-preserving DHR framework called DP-DHR. Gaussian random noise is injected into each (online) executor output in DP-DHR to guarantee DP, but it also makes the decision module unable to choose the final result because each executor output is potentially correct even if it is compromised by adversaries. To weaken this disadvantage, we propose the advanced decision strategy and the hypersphere clustering algorithm to classify the perturbed intermediate results into two categories, candidates and outliers, where the former is closer to the correct value than the latter. Finally, the DP-DHR is proven to guarantee DP, and the experimental results also show that the utility is not sacrificed for the enhancement of privacy by much (a ratio of 4–7% on average), even in the condition of some executors (less than one-half) being controlled by adversaries.

show abstract

Section: Remarkmentioning

confidence: 99%

A Differentially Private Framework for the Dynamic Heterogeneous Redundant Architecture System in Cyberspace

Kang,

Zhang,

Jiang

et al. 2024

Electronics

View full text Add to dashboard Cite

show abstract

“…All three metrics (DM, accuracy, and PD) were used to measure the effectiveness of the proposed approach. For fair comparisons, we prepared anonymized versions of both datasets with varying scales of k. We used both large-scale values (L s , where L s = D k and k = [50,75,100,125,150,175,200,250]), and small-scale values (S s , where S s = D k and k = [5,10,15,20,25,30,35,40]) of k to demonstrate the potency of the proposed approach.…”

Section: Accuracy =mentioning

confidence: 99%

“…DP [24] and its improved versions are mostly used in this mechanism of PPDP. Although DP provides strong privacy guarantees, the utility of the resulting dataset is often low, especially when a small is used [25,26]. Furthermore, the amount of noise injected by the DP model in less frequent parts of data is very high, leading to poor utility in data-driven applications.…”

mentioning

confidence: 99%

A Generic Approach towards Enhancing Utility and Privacy in Person-Specific Data Publishing Based on Attribute Usefulness and Uncertainty

Majeed

Hwang

2023

Electronics

View full text Add to dashboard Cite

This paper proposes a generic anonymization approach for person-specific data, which retains more information for data mining and analytical purposes while providing considerable privacy. The proposed approach takes into account the usefulness and uncertainty of attributes while anonymizing the data to significantly enhance data utility. We devised a method for determining the usefulness weight for each attribute item in a dataset, rather than manually deciding (or assuming based on domain knowledge) that a certain attribute might be more useful than another. We employed an information theory concept for measuring the uncertainty regarding sensitive attribute’s value in equivalence classes to prevent unnecessary generalization of data. A flexible generalization scheme that simultaneously considers both attribute usefulness and uncertainty is suggested to anonymize person-specific data. The proposed methodology involves six steps: primitive analysis of the dataset, such as analyzing attribute availability in the data, arranging the attributes into relevant categories, and sophisticated pre-processing, computing usefulness weights of attributes, ranking users based on similarities, computing uncertainty in sensitive attributes (SAs), and flexible data generalization. Our methodology offers the advantage of retaining higher truthfulness in data without losing guarantees of privacy. Experimental analysis on two real-life benchmark datasets with varying scales, and comparisons with prior state-of-the-art methods, demonstrate the potency of our anonymization approach. Specifically, our approach yielded better performance on three metrics, namely accuracy, information loss, and disclosure risk. The accuracy and information loss were improved by restraining heavier anonymization of data, and disclosure risk was improved by preserving higher uncertainty in the SA column. Lastly, our approach is generic and can be applied to any real-world person-specific tabular datasets encompassing both demographics and SAs of individuals.

show abstract

“…However, current research shows that calculating a posteriori the achieved privacy level is still important in order to audit the DP application process because hyperparameter tuning is complex and may result in unwanted and unnoticed errors. That is why there are methods to calculate approximate lower and higher limits of the proposed privacy level [101,[125][126][127][128]. For example, Florian Tramèr et al debugged a recently proposed DP training mechanism, proving with ∼99.9% confidence that the proposed method reaches not the proposed ε = 0.21 lower bound privacy level but rather ε > 2.79.…”

mentioning

confidence: 99%

“…With this, the goal is to decrease the necessary time and computing resources in order to find the target privacy-utility trade-off using empirical risk minimization (ERM). However, such research is limited to specific use cases, e.g., objective function perturbation [125,129]. For this reason, implementers could use more abstract guidelines.…”

mentioning

confidence: 99%

Distributed Learning in the IoT–Edge–Cloud Continuum

Arzovs,

Judvaitis,

Nesenbergs

et al. 2024

MAKE

View full text Add to dashboard Cite

The goal of the IoT–Edge–Cloud Continuum approach is to distribute computation and data loads across multiple types of devices taking advantage of the different strengths of each, such as proximity to the data source, data access, or computing power, while mitigating potential weaknesses. Most current machine learning operations are currently concentrated on remote high-performance computing devices, such as the cloud, which leads to challenges related to latency, privacy, and other inefficiencies. Distributed learning approaches can address these issues by enabling the distribution of machine learning operations throughout the IoT–Edge–Cloud Continuum by incorporating Edge and even IoT layers into machine learning operations more directly. Approaches like transfer learning could help to transfer the knowledge from more performant IoT–Edge–Cloud Continuum layers to more resource-constrained devices, e.g., IoT. The implementation of these methods in machine learning operations, including the related data handling security and privacy approaches, is challenging and actively being researched. In this article the distributed learning and transfer learning domains are researched, focusing on security, robustness, and privacy aspects, and their potential usage in the IoT–Edge–Cloud Continuum, including research on tools to use for implementing these methods. To achieve this, we have reviewed 145 sources and described the relevant methods as well as their relevant attack vectors and provided suggestions on mitigation.

show abstract

Towards practical differential privacy in data analysis: Understanding the effect of epsilon on utility in private ERM

Cited by 8 publications

References 2 publications

A Differentially Private Framework for the Dynamic Heterogeneous Redundant Architecture System in Cyberspace

A Differentially Private Framework for the Dynamic Heterogeneous Redundant Architecture System in Cyberspace

A Generic Approach towards Enhancing Utility and Privacy in Person-Specific Data Publishing Based on Attribute Usefulness and Uncertainty

Distributed Learning in the IoT–Edge–Cloud Continuum

Contact Info

Product

Resources

About