Quantum Federated Learning

Federated learning enables model training across distributed data sources without centralising raw data, offering a privacy-preserving alternative to aggregated training. However, its security guarantees are subject to attack: gradient inversion attacks can recover training data from model updates, and model poisoning by malicious participants can corrupt global model quality. Separately, quantum computing introduces both new threats — quantum attacks on classical cryptographic schemes used to protect federated protocols — and new opportunities, such as quantum-enhanced learning models and quantum-secure aggregation primitives.

This project investigates the intersection of quantum computation and federated learning, examining quantum-secure protocol design for distributed model training, quantum model architectures and their trainability, and the implications of quantum computation for the confidentiality and integrity guarantees of federated learning systems.