Client Dropout

Client dropout, the unpredictable unavailability of participating devices in federated learning (FL), significantly hinders the convergence and accuracy of collaboratively trained models. Current research focuses on mitigating this issue through algorithmic improvements, such as employing techniques that mimic missing updates or substitute them with data from similar clients, and by developing robust algorithms that are less sensitive to client participation variability. These efforts aim to improve the reliability and efficiency of FL, particularly in resource-constrained environments like mobile edge networks, and are crucial for expanding the practical applications of this privacy-preserving machine learning paradigm.

Papers

December 16, 2024

UA-PDFL: A Personalized Approach for Decentralized Federated Learning
Hangyu Zhu, Yuxiang Fan, Zhenping Xie
Decentralized Federated Learning Machine Learning Paradigm Personalization Layer Client Data Client Dropout

June 21, 2023

MimiC: Combating Client Dropouts in Federated Learning by Mimicking Central Updates
Yuchang Sun, Yuyi Mao, Jun Zhang
Loss Function Pre Change Information FedAvg Algorithm Novel Mimic Algorithm Client Dropout Stable Convergence

May 16, 2023

Keep It Simple: Fault Tolerance Evaluation of Federated Learning with Unreliable Clients
Victoria Huang, Shaleeza Sohail, Michael Mayo, Tania Lorido Botran, Mark Rodrigues, Chris Anderson, Melanie Ooi
Model Training Fault Tolerance Client Dropout

March 11, 2023

Stabilizing and Improving Federated Learning with Non-IID Data and Client Dropout
Jian Xu, Meiling Yang, Wenbo Ding, Shao-Lun Huang
Raw Data Label Distribution Biased Classifier Classification Head Client Dropout

August 14, 2022

Long-Short History of Gradients is All You Need: Detecting Malicious and Unreliable Clients in Federated Learning
Ashish Gupta, Tie Luo, Mao V. Ngo, Sajal K. Das
Long Span Defense Algorithm Malicious Client Client Dropout

July 18, 2022

Study of the performance and scalability of federated learning for medical imaging with intermittent clients
Judith Sáinz-Pardo Díaz, Álvaro López García
Deep Learning LeArning Abstract System Performance Medical Image Analysis Medical Imaging Potential Scalability Decentralized Data Client Dropout

May 26, 2022

Combating Client Dropout in Federated Learning via Friend Model Substitution
Heqiang Wang, Jie Xu
Client Participation Substitute Model Partial Participation Client Dropout

Client Dropout

Papers

UA-PDFL: A Personalized Approach for Decentralized Federated Learning

MimiC: Combating Client Dropouts in Federated Learning by Mimicking Central Updates

Keep It Simple: Fault Tolerance Evaluation of Federated Learning with Unreliable Clients

Stabilizing and Improving Federated Learning with Non-IID Data and Client Dropout

Long-Short History of Gradients is All You Need: Detecting Malicious and Unreliable Clients in Federated Learning

Study of the performance and scalability of federated learning for medical imaging with intermittent clients

Combating Client Dropout in Federated Learning via Friend Model Substitution