What Is Federated Learning?
Federated learning is a machine learning technique where AI models are trained across multiple decentralised devices or servers — such as smartphones, hospital systems, or bank branches — without exchanging the underlying raw data. Each device trains a local version of the model on its own data and shares only the model updates (weights, gradients) with a central server, which aggregates these updates to improve the global model. The raw data never leaves the local device. This approach supports data minimisation and privacy-by-design principles because personal data stays on the device where it was collected.
Why it matters for governance
While federated learning enhances privacy, it creates distinct governance challenges. Model convergence may be uneven across devices with different data distributions, leading to performance disparities across user groups. Model updates (gradients) can potentially be reverse-engineered to infer information about the training data (gradient inversion attacks). The quality and representativeness of data across participating devices is difficult to monitor centrally. Governance should address fairness across different data distributions, security of model update transmission, participant consent and withdrawal rights, and monitoring of aggregated model performance.