Running a Federation (MNIST Example)

We will be training an MNIST classifier using federated learning and two collaborators. We will use an flplan (keras_cnn_mnist_2.yaml) already provided in the repo inside /bin/federations/plans, as well as a provided collaborators list file (cols_2.yaml inside /bin/federations/collaborator_lists) containing two collaborator names, ‘col_0’ and ‘col_1’. Both collaborator names are already provided in the default local_data_config file so that the framework can locate the collaborator specific data information (which in this case consists of a shard number to use for hard-coded sharding logic that is performed after grabbing MNIST from a hard coded online location).

• Setup and Configuration
  • On the Aggregator
  • On each Collaborator Running on a Different Machine than the Aggregator
• Running the Federation
  • On the Aggregator
  • On each Collaborator [col_number = ‘0’ and ‘1’]: