Batteries, solar inverters, GPU clusters and grid controllers are increasingly managed by machine-learning models that learn from many sites at once. A successful attack on that learning, or on the telemetry it depends on, can cause physical damage rather than just data loss.
This direction applies RISE’s work on robust and privacy-preserving federated learning, secure communication and continuous monitoring to infrastructure that has to keep running.
Current work
- Privacy-preserving federated learning for battery health estimation across distributed energy storage assets, funded through the TRaCE Lab to Market project.
- Software-interface security for battery management systems and trustworthy, energy-aware scheduling of residential GPU networks.
- Secure communication and telemetry integrity so that operational decisions rest on data that has not been tampered with.
- Fairness-aware and privacy-preserving federated learning for critical-infrastructure protection, and defences against the poisoning and inference attacks that target it.