First, I would like to thank the authors for their open source work. And can you provide the core code for the RCL loss? Or indicate exactly where it is in the repository, as I didn't find where the core code is after skimming.
Thanks for your interest. We just refactored some things for easier understanding.
RCLClient manages client-side operations for FedRCL traning.
CLLoss is the implementation of contrastive learning loss. We implement RCL loss by a summation of SCL loss and divergence penalty loss as in here, each of which is a variant of CL loss.
fedrcl.yaml shows the detailed configurations of RCL Loss. You can directly change the values here, or can create the new yaml files.
First, I would like to thank the authors for their open source work. And can you provide the core code for the RCL loss? Or indicate exactly where it is in the repository, as I didn't find where the core code is after skimming.