ReFedEz: A federated learning framework easy to adapt, deploy and reproduce

ReFedEz is a federated learning framework designed to simplify deployment while enforcing reproducible execution environments and encrypted communication between nodes. Traditional federated learning deployments often assume environmental homogeneity while operating in heterogeneous infrastructures, introducing variance that compromises reproducibility and complicates experimental validation. Rather than prioritizing extensibility, ReFedEz focuses on the requirements necessary to guarantee controlled and auditable training conditions, particularly in regulated sectors. The framework builds and distributes immutable, version-pinned execution environments across all participants, ensuring that experiments correspond to identical runtime conditions. It manages node coordination and supports common machine learning libraries such as TensorFlow, PyTorch, and NumPy.