CCST
CCST copied to clipboard
feddg (ELCFS) meta learning implementation is wrong
https://github.com/JeremyCJM/CCST/blob/87c3e448cfa28c366b58d17f8dffc5d3d83574e5/federated/fed_run.py#L112-L122
Your meta learning implementation is wrong, load_state_dict() does not preserve the computation graph so loss_outer will never propagate gradient back to the original model parameters. be careful :)