ACM MM

FedBCD:Communication-Efficient Accelerated Block Coordinate Gradient Descent for Federated Learning

FedBCD：用于联邦学习的通信高效加速块坐标梯度下降法

Junkang Liu, Fanhua Shang, Yuanyuan Liu, Hongying Liu, Yuangang Li, YunXiang Gong

机构 * School of Artificial Intelligence Xi'an Xidian University, China ； College of Intelligence ； Computing Tianjin Tianjin University, China ； Medical College, Tianjin University\ Cheng Laboratory Tianjin China ； University of Southern California Los Angeles US ； School of Artificial Intelligence ； Medical College, Tianjin University\ Cheng Laboratory ； University of Southern California 纠错

AI总结 本文提出FedBCGD方法，通过划分模型参数为多个块以降低通信开销，并开发FedBCGD+算法实现加速，首次在大规模深度模型中应用参数块通信。

Journal ref ACM MM 2024

FedNSAM:Consistency of Local and Global Flatness for Federated Learning

FedNSAM：联邦学习中局部与全局平坦度的一致性

Junkang Liu, Fanhua Shang, Yuxuan Tian, Hongying Liu, Yuanyuan Liu

机构 * College of Intelligence ； Computing, Tianjin University Tianjin China ； College of Management ； Economics, Tianjin University Tianjin China ； Medical College, Tianjin University\ Cheng Laboratory Tianjin China ； School of Artificial Intelligence, Xidian University Xian China ； Computing, Tianjin University ； Economics, Tianjin University ； Medical College, Tianjin University\ Cheng Laboratory ； School of Artificial Intelligence, Xidian University 纠错

AI总结 FedNSAM通过引入全局Nesterov动量提升SAM效果，解决高数据异质性下局部平坦度不保证全局模型平坦度的问题，理论证明收敛界更紧，实验验证性能优越。

Journal ref ACM MM 2025