FairMean: Promoting Fairness in Distributed Learning under Label Poisoning Attacks
Experiments show that FairMean promotes fairness by reducing accuracy variance while improving worst-client accuracy, and it is shown that minimizing the FairMean objective is more conducive to solution fairness than minimizing the standard average-loss objective.
Hui-Gan Zheng, Jiao-Jiao Zhang, Yong-Xiang Liu
· 0 citations