Privacy-preserving measurement and machine learning

Cloudflare has developed a protocol called Prio that enables privacy-preserving measurement of user data while maintaining their privacy. The process involves secret sharing and special zero-knowledge proofs to ensure no single bit of the measurements can be discerned by an aggregator, even during interaction for non-linear computation tasks like finding heavy hitters among the set of measurements. The PPM working group at IETF aims to standardize multi-party computation (MPC) techniques for privacy preserving measurement through three main prongs: identifying problems that need solutions, providing cryptography researchers with "templates" such as Verifiable Distributed Aggregation Functions (VDAFs), and offering a deployment roadmap for emerging protocols like Prio. Cloudflare has also developed an implementation of a DAP aggregator server called Daphne using Rust and Workers, which allows easy integration with their storage solutions like Durable Objects and KV. The company is looking to collaborate on the project or participate in the development of MPC techniques for privacy-preserving measurement through the PPM working group at IETF. Cloudflare's contributions have helped improve both their Workers products and the DAP standard, as well as provide feedback into the protocol's interoperability tests during deployment.