FRIDGE (Federated Research Infrastructure by Data Governance Extension)

Introduction

To apply cutting edge artificial intelligence (AI) models to problems that require access to sensitive data we need to be able to use this sensitive data on large-scale shared AI supercomputers, like the UK’s new AI Research Resource (AIRR). This requires the ability to strongly isolate parts of these shared system as Trusted Research Environments (TREs) so that access to the sensitive data is restricted to those researchers approved to work with it.

Currently there is no such capability; FRIDGE aims to solve that.

Explaining the science

The FRIDGE team consists of experts in building TREs (The Alan Turing Institute and University College London) as well as infrastructure engineers working at the two AIRR sites, Dawn and Isambard-AI. This team will build a TRE system that can run on either AIRR platform and be easily re-used by other organisations wanting to do large scale AI research using sensitive data. Designing the system in this platform-agnostic way will make it possible to deploy the same TRE system on other computing platforms than AIRR, extending the range of computing resources available to support research using sensitive data.

Project aims

FRIDGE will deliver a portable, cross-platform SATRE and NHS standards compliant Trusted Research Environment (TRE) into production on AIRR, the UK’s national AI research supercomputing platform. This will provide the ability to strongly isolate parts of the AIRR system for an individual project that uses sensitive data, so that this data can only be accessed by approved researchers and can be administered under the information governance regime approved for the project. The software and associated legal agreements and policies for systems like AIRR to run the FRIDGE TRE will be published openly for others to easily reuse, either to deploy their sensitive data projects to the AIRR platform or to enable other supercomputers to support hosting such projects in TREs on their systems.

Applications

FRIDGE unlocks the potential of using high-performance computing (HPC) systems for any research project requiring the use of sensitive data. The ability to run a TRE in an isolated section of a HPC system means that control of the data remains in the hands of the research institution who has been entrusted with it to carry out the approved research. This greatly simplifies data governance, allowing any organisation to extend their existing processes to cover this new compute resource without requiring the organisation running the HPC system to become a data controller.

The open-source FRIDGE codebase will also be platform agnostic, enabling any organisation to adapt it for their own use, without the need to develop their own system from scratch.