The minimax problems arise throughout machine learning applications, ranging
from adversarial training and policy evaluation in reinforcement learning to
AUROC maximization. To address the large-scale data challenges across multiple
clients with communication-efficient distributed trai