This research project aims to explore how federated learning can be deployed to support responsible data stewardship and ensure that data is made available to address the critical challenges of our time. Our objective is to create accessible guidance for organisations seeking to deploy federated learning. We hope that the introduction of the concepts of data stewardship and data institutions can inspire developers and practitioners of federated learning techniques to apply them towards public, educational and charitable purposes.
The Open Data Institute (ODI)’s work on data institutions has highlighted the vital role these organisations play by ensuring that data is stewarded – collected, maintained and shared – to pursue public, charitable or educational aims.
Organisations that steward data make important decisions about who has access to it, for what purposes and to whose benefit. How data is stewarded ultimately affects what types of products, services and insights it can be used to create, what decisions it can inform and which activities it can support. Stewarding data involves realising the value and limiting the harm that data can bring.
One important way in which data institutions steward data is by facilitating access to sensitive data in safe, ethical and equitable ways. There is value in increasing access to sensitive data, such as data about health, transport or demographics, but it must be done in ways that respect things like privacy, commercial sensitivity and national security.
Federated learning, a machine learning technique that trains an algorithm across data from multiple devices, has emerged as a promising new approach to access and analyse data. By enabling algorithms to be trained across multiple local datasets without exchanging the underlying data, it presents great potential to unlock value from data that might traditionally have been kept closed.
There are some examples of federated learning being trialled within the healthcare sector – for instance, the EU’s MELLODDY project – and there is also potential for the application of this technology in other sectors and domains.
Between May and December 2022, the ODI will conduct desk research, expert interviews and analysis to understand how organisations can use federated learning as a technique to steward data more responsibly and effectively. The project is supported by the Rockefeller Foundation, and will inform the Foundation and Data.org’s wider work on Building data science innovations that drive social impact and the The Epiverse: Distributed Pandemic Tools Program.
Get in touch
Given the relative novelty of federated learning, engaging with stakeholders who are currently in the process of developing and deploying the technology is a vital part of our process. If you are working on federated learning techniques – particularly for public, educational and charitable purposes – we would love to hear from you. Please get in contact with us at [email protected]