[Paper Review] Advancing Differential Privacy: Where We Are Now and Future Directions for Real-World Deployment

Rachel Cummings, Damien Desfontaines+22 more|arXiv (Cornell University)|Apr 14, 2023

Privacy, Security, and Data Protection9 citations

TL;DR

A workshop-derived overview of challenges, tooling needs, and research directions for deploying differential privacy in industry-scale systems, with emphasis on infrastructure, privacy/utility trade-offs, auditing, and communication of guarantees.

ABSTRACT

In this article, we present a detailed review of current practices and state-of-the-art methodologies in the field of differential privacy (DP), with a focus of advancing DP's deployment in real-world applications. Key points and high-level contents of the article were originated from the discussions from "Differential Privacy (DP): Challenges Towards the Next Frontier," a workshop held in July 2022 with experts from industry, academia, and the public sector seeking answers to broad questions pertaining to privacy and its implications in the design of industry-grade systems. This article aims to provide a reference point for the algorithmic and design decisions within the realm of privacy, highlighting important challenges and potential research directions. Covering a wide spectrum of topics, this article delves into the infrastructure needs for designing private systems, methods for achieving better privacy/utility trade-offs, performing privacy attacks and auditing, as well as communicating privacy with broader audiences and stakeholders.

Motivation & Objective

• Map the main practical challenges in deploying differential privacy (DP) in industry and public sector contexts.
• Identify desiderata for DP tooling and infrastructure to support adoption by non-experts.
• Outline research directions to improve privacy/utility trade-offs and scalability of DP algorithms and systems.
• Discuss methods for auditing, attacks, and benchmarking to validate DP implementations.
• Explore how to communicate DP guarantees to end-users and stakeholders while considering real-world constraints.

Proposed method

• Synthesize insights from a July 2022 DP workshop with industry, academia, and public-sector participants.
• Detail practical challenges in deployment and propose research agendas for tooling, benchmarking, and usability.
• Provide guidance on implementing DP in end-to-end systems, including threat models and composition considerations.
• Discuss public data usage, data minimization, and infrastructure design to improve DP adoption and effectiveness.

Experimental results

Research questions

• RQ1 What are the key challenges and barriers to real-world deployment of differential privacy across industries?
• RQ2 What tooling, infrastructure, and process improvements are needed to support scalable and usable DP systems?
• RQ3 How can DP trade-offs be improved through public data, data-adaptive methods, and DP-enabled pipelines?
• RQ4 How can we effectively audit, attack-test, and benchmark DP deployments to validate guarantees?
• RQ5 How should DP guarantees be communicated to end-users and stakeholders in practice?

Key findings

• DP has become the dominant privacy notion for statistical analysis and ML, but real-world adoption faces challenges in threat modeling, parameter selection, integration, and cost.
• Robust DP tooling should focus on usability, trust through open-source implementations, support for diverse privacy units, and scalability to large datasets.
• Public data and data-adaptive/transfer learning approaches show promise for improving DP utility, especially in pretraining and private fine-tuning of models.
• Auditing and attacks are valuable to bound information leakage and guide parameter choices, though they do not replace worst-case guarantees and must be interpreted carefully.
• Effective privacy communication requires accounting for DP as part of a broader privacy strategy including data minimization and transparency, with attention to policy and regulatory considerations.