Hermes Paper Reading Notes
Published:
Hermes: An Efficient Federated Learning Framework for Heterogeneous Mobile Clients
Overview
Step
- Each device incorporates a structured sparsity regularization to learn a subnetwork when optimizing the local model.
- Only such subnetworks are transmitted from devices to the central server.
- In the central server, only the parameters are intersected across the subnetworks of devices are averaged while keeping the remaining non-intersected parameters untouched.
- The updated subnetworks will be distributed to each device.
- Repeat steps 1~4 until reaching the predefined number of communication rounds.
- Each devices will finally obtain a personalized and structured-sparse model.