Federated learning for aircraft engine prognostics

A new federated learning study addresses benign and adversarial heterogeneity in aircraft engine sensor telemetry. Using a multi-task 1D-CNN on the C-MAPSS benchmark, the method uses shared-representation personalization to close 70% of the error gap between local and centralized models.