search

yuetan031 / FedPCL

[NeurIPS'22 Spotlight] Federated Learning from Pre-Trained Models: A Contrastive Learning Approach
40 stars 8 forks source link

Global prototype-based loss computing is missing #3

Open TsingZ0 opened 1 year ago

TsingZ0 commented 1 year ago

I cannot find the global prototype-based loss computing in update_weights_lg(). Line 162 actually computes the local prototype-based loss

TsingZ0 commented 1 year ago

Besides, the inference is also not done with local prototypes in Line 224.