Differential Privacy

1 post · Sep 2023 – Dec 2023

What this trail covers

Cloudflare is actively involved in standardizing and implementing privacy-preserving technologies. This includes contributing to the Distributed Aggregation Protocol (DAP) for private aggregation and enhancing it with differential privacy to provide a stronger guarantee against data leakage. The company is developing an open-source implementation of DAP and a service to support partners, aiming to protect user privacy in various applications like telemetry and exposure notification systems.

All Cloudflare trails

Dec 2023Have your data and hide it too: an introduction to differential privacyThis post introduces differential privacy as a method to enhance the privacy guarantees of the Distributed Aggregation Protocol (DAP). It explains the shortcomings of private aggregation alone, illustrating how aggregates can still reveal sensitive information through attacks like those on the U.S. Census and LLMs. Differential privacy is presented as a statistical framework that adds noise to aggregates to protect individual data, with the privacy parameter 'epsilon' quantifying the information revealed. The post highlights the benefits of DP, including its generality, harmonized security parameters, and graceful degradation properties compared to ad-hoc methods. It also mentions an internship project to devise a strategy for endowing DAP with differential privacy.
Sep 2023Privacy-preserving measurement and machine learningThis post details the fundamental concepts behind the Distributed Aggregation Protocol (DAP), a standard for privacy-preserving measurement. It explains how secret sharing and multi-party computation (MPC) can be used to aggregate data without exposing individual measurements. The post also introduces Daphne, Cloudflare's open-source aggregator server implementing DAP, and discusses the challenges and solutions for input validation using zero-knowledge proofs (ZKPs) and handling non-linear aggregation tasks.
Advertisement