Creating a Decentralised federated learning scenarios in tensorflow

Hello Everyone

I am trying to create a Decentralized federated learning scenario in TensorFlow, where the clients communicate among themselves and update their parameters; there is no central server or an aggregator like in the case of centralized federated learning.

  1. Is it possible to create this scenario in TensorFlow
  2. Is it possible to use lambda functions as a ping to do the parameter updates among the client network
  3. I have my custom pre-trained models in PyTorch can I transfer them to TensorFlow and use them here, if so how can I export my models to TensorFlow

If there are any other suggestions that would help me build this scenario please let me know

1 Like

Iā€™m hoping to do the same. @saras26 were you able to gain any new insight on this?