FedMM: Saddle Point Optimization for Federated Adversarial Domain Adaptation
This addresses the challenge of training global models in federated settings with domain-shifted and imbalanced data, offering a novel optimization approach for this specific problem.
The paper tackles the problem of federated adversarial domain adaptation with label imbalance among clients by proposing FedMM, a distributed minimax optimizer. It shows that FedMM achieves significant improvements, such as around 20% higher accuracy over gradient descent ascent methods in some cases.
Federated adversary domain adaptation is a unique distributed minimax training task due to the prevalence of label imbalance among clients, with each client only seeing a subset of the classes of labels required to train a global model. To tackle this problem, we propose a distributed minimax optimizer referred to as FedMM, designed specifically for the federated adversary domain adaptation problem. It works well even in the extreme case where each client has different label classes and some clients only have unsupervised tasks. We prove that FedMM ensures convergence to a stationary point with domain-shifted unsupervised data. On a variety of benchmark datasets, extensive experiments show that FedMM consistently achieves either significant communication savings or significant accuracy improvements over federated optimizers based on the gradient descent ascent (GDA) algorithm. When training from scratch, for example, it outperforms other GDA based federated average methods by around $20\%$ in accuracy over the same communication rounds; and it consistently outperforms when training from pre-trained models with an accuracy improvement from $5.4\%$ to $9\%$ for different networks.