Privacy-preserving data analysis

Introduction

Huge amounts of data exist about every one of us, the use of which has the potential to improve our lives and the world we live in. However, concerns about the privacy of this data have naturally become an increasingly prevalent issue. The aim of privacy-preserving analysis is to utilise this data to its fullest potential without compromising our privacy.

Explaining the science

In this talk we consider FHE schemes based on the Ring Learning with Errors (Ring-LWE) problem and focus on two matters of practical interest. The first is the selection of secure Ring-LWE parameters. In particular, we discuss the differing estimates used in the submissions to the NIST process for the running time of algorithms to solve Ring-LWE. The second is the choice of an appropriate encoding from raw data into the plaintext space, which can impact on efficiency.

Privacy-preserving algorithms for decentralised collaborative learning: Dr Aurélien Bellet

Dr Emiliano De Cristofaro's research is in security and privacy enhancing technologies. He's currently working on understanding and countering security issues via measurement studies and data-driven analysis, as well as tackling problems at the intersection of machine learning and security & privacy.

Private set intersection (PSI) allows two parties to compute the intersection of their sets without revealing any information about items that are not in the intersection. This talk surveys several custom PSI protocols, and describe how to apply generic MPC protocols to computing PSI while computing only a linear number of comparisons.

Differential privacy is a robust mathematical framework for designing privacy-preserving computations on sensitive data. This tutorial covers the key definitions and intuitions behind differential privacy and introduces the core building blocks used by most differentially private mechanisms.

Hiding memory access patterns is required for secure computation, but remains prohibitively expensive for many interesting applications. This talk presents two works addressing this question: a new oblivious RAM (ORAM) construction and a secure computation scheme using ORAM in the context of Boolean database queries.

Aims

To understand the interplay between different privacy-enhancing techniques and how they can be used in practice for privacy-preserving data analysis.

It is important to develop a unified approach to secure, privacy-preserving data analysis as well as finding an effective, mathematically robust definition of privacy.

We will organise periodic workshops and talks at the Turing, as well as lectures and tutorials aimed at a general audience. Although the focus of the group is on technical aspects, engaging with researchers on ethical and regulatory aspects will be one of the workshops’ goals.

Why now?

Privacy-preserving data analysis has become a crucial aspect of data science, and is recognised as an important problem within several research communities.
Recent advances in cryptography, systems, and hardware security, have made privacy-preserving computation practical.
There are several deployments in existing and new products, and lots of interest both from industry and the government.