[Paper Review] Local Differential Privacy and Its Applications: A Comprehensive Survey

Mengmeng Yang, Lingjuan Lyu+3 more|arXiv (Cornell University)|Aug 9, 2020

Privacy-Preserving Technologies in Data138 references65 citations

TL;DR

This paper surveys Local Differential Privacy (LDP) techniques and applications, covering frequency estimation, heavy hitters identification, and set-valued data, with comparisons of major perturbation mechanisms and design trade-offs.

ABSTRACT

With the fast development of Information Technology, a tremendous amount of data have been generated and collected for research and analysis purposes. As an increasing number of users are growing concerned about their personal information, privacy preservation has become an urgent problem to be solved and has attracted significant attention. Local differential privacy (LDP), as a strong privacy tool, has been widely deployed in the real world in recent years. It breaks the shackles of the trusted third party, and allows users to perturb their data locally, thus providing much stronger privacy protection. This survey provides a comprehensive and structured overview of the local differential privacy technology. We summarise and analyze state-of-the-art research in LDP and compare a range of methods in the context of answering a variety of queries and training different machine learning models. We discuss the practical deployment of local differential privacy and explore its application in various domains. Furthermore, we point out several research gaps, and discuss promising future research directions.

Motivation & Objective

• Survey the landscape of Local Differential Privacy (LDP) techniques and their applicability to real-world data collection.
• Analyze and compare perturbation mechanisms (e.g., GRR, OUE, RAPPOR, OLH, JLRR, HRR) for frequency estimation under LDP.
• Examine heavy hitters identification methods under LDP and discuss efficiency, communication, and accuracy trade-offs.
• Discuss set-valued data collection under LDP and the challenges of heterogeneous set sizes.
• Summarize open challenges and practical guidance for choosing LDP techniques under different privacy-budget regimes.

Proposed method

• Describe transformation-based methods that convert user data into compact representations to reduce communication while preserving privacy.
• Present subset selection and partition/tree-based approaches to identify heavy hitters efficiently under LDP.
• Summarize protocol variants for set-valued data collection including trimming/padding and domain size handling.
• Provide comparisons of typical heavy hitter and frequency estimation protocols (GRR, OUE, RAPPOR, OLH, JLRR, HRR) in terms of variance, communication, and domain size.
• Discuss algorithmic considerations such as bias, variance, and the impact of privacy budget ε on statistical accuracy.

Experimental results

Research questions

• RQ1 What are the main LDP mechanisms for frequency estimation and how do they compare in terms of variance and communication cost?
• RQ2 How can heavy hitters be efficiently identified under LDP, and what are the trade-offs of partition-based, tree-based, and naive approaches?
• RQ3 How does set-valued data collection operate under LDP, and what are the challenges with heterogeneous set sizes?
• RQ4 What are the practical considerations and guidelines for selecting LDP techniques across low, medium, and high privacy budgets?
• RQ5 What open challenges remain in balancing privacy, accuracy, and efficiency in LDP applications?

Key findings

• Frequency estimation under LDP relies on several core perturbation methods (GRR, OUE, RAPPOR, OLH, JLRR, HRR) with distinct variance and communication characteristics.
• Heavy hitters identification under LDP can be approached via naive, partition-based, and tree-based methods, each with specific efficiency, accuracy, and communication trade-offs.
• Partition-based and tree-based methods improve efficiency for high-dimensional data but introduce accuracy and communication challenges, especially as the domain size grows or privacy budget is small.
• Set-valued data collection under LDP faces heterogeneity in set sizes; methods trim/pad and use padding to enable frequency estimation and itemset mining.
• Transformations and subset selection can reduce communication costs, but may increase statistical variance or require more complex coordination.
• Overall, frequency estimation is the most mature area under LDP, with ongoing challenges in high-dimensional data and balancing privacy with accuracy and communication constraints.