Glossary

Here below the glossary used for Fed-BioMed :

• experiment : orchestrates the rounds during the federated learning, on the available nodes

  • it includes : model, federated trainer, training parameters, model parameters, set of input data, results
  • an experiment is unique (cannot be replayed) and is over when converged
  • status : running and then done

• job : not a researcher notion. Interface between the researcher and the nodes of an experiment. It triggers the local work for all sampled nodes at each round.

• round : everything included in choice of the nodes, perform local work on the nodes, sending back whatever information is required, server performs the aggregation

  • current Round() class on node corresponds to local work

• researcher (technical) : entity that defines and executes an experiment

• node (technical) : entity with tagged datasets that replies to researcher queries and performs local work