In this work, we developed a novel algorithm named divergence-based adaptive aggregation (DRAG) to deal with the client-drift effect. Additionally, the DRAG algorithm also showcases resilience against byzantine attacks, as demonstrated through experiments.
Implementation of Local Updates Periodic Averaging (LUPA) SGD
Distributed Optimization Methods for Machine Learning
This work is a journal paper published at IEEE TSP in 2024, concentrating on improving the robustness to stragglers in distributed/federated learning with synchronous local SGD.
This work is a journal paper published at The Journal of Supercomputing in 2023, focusing on improving communication efficiency of a distributed learning system using age-based worker selection techniques.
Add a description, image, and links to the local-sgd topic page so that developers can more easily learn about it.
To associate your repository with the local-sgd topic, visit your repo's landing page and select "manage topics."