Federated Learning
联邦学习是一种机器学习方法,允许多个设备或实体在不交换数据的情况下协同训练共享模型。各设备在本地进行模型训练,仅将模型更新发送至中央服务器进行聚合,以优化共享模型。这种方法实现了隐私保护的机器学习,确保数据本地存储,仅共享必要的模型改进信息。联邦学习的目标是提升模型性能,同时保障用户数据安全与隐私,具有广泛的应用价值。
CIFAR-100 (alpha=0, 10 clients per round)
CIFAR-100 (alpha=0, 20 clients per round)
CIFAR-100 (alpha=0.5, 10 clients per round)
CIFAR-100 (alpha=0.5, 20 clients per round)
CIFAR-100 (alpha=0.5, 5 clients per round)
CIFAR-100 (alpha=0, 5 clients per round)
CIFAR-100 (alpha=1000, 10 clients per round)
CIFAR-100 (alpha=1000, 20 clients per round)
CIFAR-100 (alpha=1000, 5 clients per round)
FedASAM
CIFAR100 (alpha=0.3, 10 clients per round)
AdaBest
Cityscapes heterogeneous
Landmarks-User-160k