《计算机应用研究》|Application Research of Computers

基于类别不平衡数据联邦学习的设备选择算法

Device selection in federated learning under class imbalance

免费全文下载 (已被下载 次)  
获取PDF全文
作者 王惜民,范睿
机构 上海科技大学 信息科学与技术学院,上海 201210
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2021)10-014-2968-06
DOI 10.19734/j.issn.1001-3695.2021.03.0045
摘要 考虑移动边缘计算下的联邦学习,其中全局服务器通过网络连接大量移动设备共同训练深度神经网络模型。全局类别不平衡和设备本地类别不平衡的数据分布往往会导致标准联邦平均算法性能下降。提出了一种基于组合式多臂老虎机在线学习算法框架的设备选择算法,并设计了一种类别估计方案。通过每一轮通信中选取与前次全局模型的类别测试性能偏移最互补的设备子集,使得训练后线性组合的全局模型各类别测试性能更平衡,从而获得更快的收敛性、更稳定的训练过程以及更好的测试性能。数值实验充分探究了不同参数对基于类别不平衡联邦平均算法的影响,以及验证了所提设备选择算法的有效性。
关键词 联邦学习; 移动边缘计算; 深度学习; 组合式多臂老虎机; 隐私保护
基金项目 上海科技大学启动基金资助项目(2017F0203-000-05)
本文URL http://www.arocmag.com/article/01-2021-10-014.html
英文标题 Device selection in federated learning under class imbalance
作者英文名 Wang Ximin, Fan Rui
机构英文名 School of Information Science & Technology,ShanghaiTech University,Shanghai 201210,China
英文摘要 Considering federated learning under mobile edge computing, where a global server connects a large number of mobile devices through the network to jointly train a deep neural network model. The data distribution shift caused by global class imbalance and device local class imbalance leads to the performance degradation of the standard federated averaging(FedAvg) algorithm. This paper proposed a device selection algorithm based on the combinatorial multi-armed bandit(CMAB) as an online learning algorithm framework, and designed a class estimation scheme to form a nonlinear reward function. Based on class estimation scheme and CMAB, in each round of communication, the global server selected the device subset with class imba-lance that could be best complementary with test performance deviation across classes of global model from last communication round. Therefore, the current aggregated global model could achieve more balance and better test performance per class, and also faster convergence and more stable training dynamics. Extensive numerical results demonstrate the influence of different parameters on FedAvg under class imbalance, and verify the effectiveness of the proposed algorithm.
英文关键词 federated learning; mobile edge computing; deep learning; combinatorial multi-armed bandit; privacy protection
参考文献 查看稿件参考文献
 
收稿日期 2021/3/13
修回日期 2021/4/26
页码 2968-2973
中图分类号 TP391
文献标志码 A