You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I read ur paper and notice that you give a benchmark result on feminist dataset. U give the model of LeNet and MLP_128 results on FedScope framework. I also check the source code of FederateScope and find there's no any implement of those two models. I notice that u give the config of LeNet on Feminist dataset. So I wonder how did u test those two models? Did u implement
by hand or what?
The text was updated successfully, but these errors were encountered:
Moreover, I check FedScope and run the feminist dataset using mlp and lr, found the results are extremely different. Wonder if u can provide the original config of FedScope using these shallow layers. Thanks
I read ur paper and notice that you give a benchmark result on feminist dataset. U give the model of LeNet and MLP_128 results on FedScope framework. I also check the source code of FederateScope and find there's no any implement of those two models. I notice that u give the config of LeNet on Feminist dataset. So I wonder how did u test those two models? Did u implement
by hand or what?
The text was updated successfully, but these errors were encountered: