CDMA: A Practical Cross-Device Federated Learning Algorithm for General Minimax Problems

Minimax problems arise in a wide range of important applications including robust adversarial learning and Generative Adversarial Network (GAN) training. Recently, algorithms for minimax the Federated Learning (FL) paradigm have received considerable interest. Existing federated general require full aggregation (i.e., local model information from all clients) each training round. Thus, they are inapplicable to an setting FL known as cross-device setting, which involves numerous unreliable mobile/IoT devices. In this paper, we develop first practical algorithm named CDMA setting. is based on Start-Immediately-With-Enough-Responses mechanism, server signals subset clients perform computation then starts aggregate results reported by once it receives responses enough With resilient low client availability. addition, incorporated with lightweight global correction update steps clients, mitigates impact slow network connections. We establish theoretical guarantees under different choices hyperparameters conduct experiments AUC maximization, training, GAN tasks. Theoretical experimental demonstrate efficiency CDMA.