Server Learning Rate

Server learning rate optimization in federated learning (FL) focuses on efficiently updating the global model based on updates from decentralized clients, addressing challenges like client heterogeneity and non-independent and identically distributed (non-i.i.d.) data. Current research emphasizes adaptive learning rate mechanisms, often employing online meta-learning or techniques inspired by coding theory, to dynamically adjust the server's update step size based on factors like client data distribution consistency or model update quality. These advancements aim to improve the speed and accuracy of FL training, particularly relevant for resource-constrained environments and privacy-sensitive applications like click-through rate prediction.

Papers

January 25, 2023

When to Trust Aggregated Gradients: Addressing Negative Client Sampling in Federated Learning
Wenkai Yang, Yankai Lin, Guangxiang Zhao, Peng Li, Jie Zhou, Xu Sun
Learning Rate Client Sampling Aggregated Gradient Server Learning Rate

August 30, 2022

Online Meta-Learning for Model Update Aggregation in Federated Learning for Click-Through Rate Prediction
Xianghang Liu, Bartłomiej Twardowski, Tri Kurniawan Wijaya
Click Through Rate Prediction Model Aggregation Click Through Rate Client Heterogeneity Online Meta Learning Server Learning Rate

January 26, 2022

Fast Server Learning Rate Tuning for Coded Federated Dropout
Giacomo Verardo, Daniel Barreira, Marco Chiesa, Dejan Kostic, Gerald Q. Maguire
Federated Learning Cross Device Federated Learning Low Computational Server Learning Rate

Server Learning Rate

Papers

When to Trust Aggregated Gradients: Addressing Negative Client Sampling in Federated Learning

Online Meta-Learning for Model Update Aggregation in Federated Learning for Click-Through Rate Prediction

Fast Server Learning Rate Tuning for Coded Federated Dropout