Fed-BioMed v4.1 is now available. Here are some key new features:

  • Introducing Scaffold Aggregation method for PyTorch, focused to cope with the client drift issue, useful when dealing with heterogeneous datasets
  • Adding num_updates as a new training_args Argument: num_updates allows you to iterate your model over a specific number of updates, regardless of the size of data across each Node. It is an alternative to number of epochs epochs
  • Adding more integration tests / introducing nightly tests in order to improve code quality
  • improving Researcher log message, by introducing Round number
  • Bug fixes (FedProx Aggregation method, percentage completion logged when using Opacus, and other minor fixes)

More details about the new features can be found in the Fed-BioMed CHANGELOG.