2026
BA-FedSHAP: A reproducible toolkit for auditing background-induced attribution drift
SSRN
Preprint on how the choice of background distribution alters federated attribution results, and what that does to an audit.
PublicationsAuthor
Research area
Privacy-preserving collaborative AI
Institutions that cannot pool sensitive data can still learn together, if the governance is built alongside the method.
- Federated learning
- Privacy
- AI assurance
What has to hold, technically and institutionally, for federated analysis across sensitive datasets to be trustworthy rather than merely decentralised?
Federated learning is often presented as a privacy guarantee. It is better understood as a change in where the risk sits. Moving computation to the data removes one exposure and introduces others: attribution that shifts with modelling choices, updates that leak, and a governance question about who is accountable when no single party holds the dataset.
I work on both halves of that, the technical behaviour of federated methods under audit, and the institutional arrangements that make a federation legible to a regulator or a ministry.
Outputs
Related work
2026
BA-FedSHAP, auditing background-induced attribution drift
Measures how much a federated explanation depends on an arbitrary modelling choice.
Research softwareAuthor
2025
Federated learning for children's data: encoding trust in education systems
UNICEF Innocenti, Office of Strategy and Evidence
Argues that governments can use sensitive education data without centralising it, and what governance that requires.
PolicyAuthor