Maybe give motivations about why the paper should be implemented as a baseline.
FedRS (Federated Learning with Restricted Softmax) is a method to correct the negative effect of local training with missing classes by restricting the update of weights via a corrected softmax term. The main parameter is alpha (between 0.0 and 1.0), which determines the strength of this correction. If alpha=1.0, this is same as vanilla FedAvg.
FedRS (100+ citations) is referenced in other papers exploring Federated Learning with non-iid datasets, such as FedLC (Federated Learning with Label Distribution Skew via Logits Calibration) and FedConcat (Exploiting Label Skews in Federated Learning with Model Concatenation).
The plan is to reproduce the results for FedAvg and FedRS (with different alpha values) in Table 5 of the paper:
Is there something else you want to add?
I've found no mention of FedRS in PRs or in existing baselines. I've started the repro process.
Implementation
To implement this baseline, it is recommended to do the following items in that order:
Paper
Xin-Chun Li, De-Chuan Zhan. FedRS: Federated Learning with Restricted Softmax for Label Distribution Non-IID Data (KDD'21)
Link
https://dl.acm.org/doi/10.1145/3447548.3467254
Maybe give motivations about why the paper should be implemented as a baseline.
FedRS (Federated Learning with Restricted Softmax) is a method to correct the negative effect of local training with missing classes by restricting the update of weights via a corrected softmax term. The main parameter is alpha (between 0.0 and 1.0), which determines the strength of this correction. If alpha=1.0, this is same as vanilla FedAvg.
FedRS (100+ citations) is referenced in other papers exploring Federated Learning with non-iid datasets, such as FedLC (Federated Learning with Label Distribution Skew via Logits Calibration) and FedConcat (Exploiting Label Skews in Federated Learning with Model Concatenation).
The plan is to reproduce the results for FedAvg and FedRS (with different alpha values) in Table 5 of the paper:
Is there something else you want to add?
I've found no mention of FedRS in PRs or in existing baselines. I've started the repro process.
Implementation
To implement this baseline, it is recommended to do the following items in that order:
For first time contributors
first contributiondocPrepare - understand the scope
Verify your implementation
EXTENDED_README.mdthat was created in your baseline directoryREADME.mdis ready to be run by someone that is no familiar with your code. Are all step-by-step instructions clear?README.mdand verify everything runs.