lxcnju / fedrepo Goto Github PK

Hi, thank you for sharing your papers' source code and implementations of several FL algorithms. I notice that you implement [FedAwS] Federated Learning with Only Positive Labels. ICML 2020. I find there is follow paper of FedAwS: Federated Learning of User Verification Models Without Sharing Embeddings (FedUV), which is published in ICML 2021. Do you have a plan to implement it?

2. And after I download the train.pkl, there is a error:
   ValueError: need at least one array to concatenate

When are you planning to publish the source code of "CVPR 2022 paper: Federated Learning with Position-Aware Neurons"?

I don't find an implementation for fedprox in the directory of `algorithms/'.

The dataset cannot be downloaded through the link given. It cannot be accessed.

我想请教一个问题

这样做是不是忽略了分类器的Bias参数呀？有什么玄机吗？

I only retrain FedAVG with default parameters (hypers['cnt']=2, C100-100-20, VGG11). And the results are $46.48+0.6703$ and $47.01+1.7480$, which is lower than the results reported in your paper: $49.8+0.471$.
Question 1: Since there are two results, the final result is the average value or the maximum value.
Question 2: The overall results are lower than those reported in the paper.