AshwinRJ
What is difference between conventional way and federated way
This is a simulation of what training using a Federated setting might look like. Since the entire code is only run on a single kernel, it is only an experimental representation of Federated Learning.
It simulates a client instance with it's own shard of data and creates an object that can take a global model and train it on it's own data shard and returns updated model weights to the simulated server.
This codebase also contains an implementation of Federated Averaging which takes multiple weights of the same model and performs layer-by-layer averaging.
In this project, federated learning only selects a portion of users in each epoch to participate in the model parameter update process, whereas in the original scheme, all users need to participate in this process
