[CVPR 2023] Federated Domain Generalization with Generalization Adjustment - presentation video
Poster Tag: TUE-AM-377
(Unfortunately, my application is still pending and there is a high likelihood that it will not be approved in time for me to attend the event in person.)
这是一篇关于隐私保护机器学习中联邦域泛化的论文。我们提出了一种新的优化目标和FL友好方法，名为Generalization Adjustment（GA），以解决现有方法无法处理公平性约束的问题。我们在多个基准数据集上进行了实验，证明了GA在提高准确性、公平性和对领域转移的鲁棒性方面优于现有FedDG方法。
This is a paper on Federated Domain Generalization in privacy-preserving machine learning. We propose a new optimization objective and FL-friendly method named Generalization Adjustment (GA) to address the limitations of existing approaches in handling fairness constraints. We conduct experiments on multiple benchmark datasets, demonstrating that GA outperforms existing FedDG methods regarding accuracy, fairness, and robustness to domain shift.
- GitHub repo: https://github.com/MediaBrain-SJTU/FedDG-GA
