Event-Based Federated Q-Learning

Abstract

This paper introduces an event-based communication mechanism in federated Q-learning algorithms, enhancing convergence and reducing communication overhead. We present a communication scheme, which leverages event-based communication to update Q-tables between agents and a central server. Through theoretical analysis and empirical evaluation, we demonstrate the convergence properties of event-based QAvg, highlighting its effectiveness in federated reinforcement learning settings.

Cite

Text

Er and Muehlebach. "Event-Based Federated Q-Learning." ICML 2024 Workshops: RLControlTheory, 2024.

Markdown

[Er and Muehlebach. "Event-Based Federated Q-Learning." ICML 2024 Workshops: RLControlTheory, 2024.](https://mlanthology.org/icmlw/2024/er2024icmlw-eventbased/)

BibTeX

@inproceedings{er2024icmlw-eventbased,
  title     = {{Event-Based Federated Q-Learning}},
  author    = {Er, Guner Dilsad and Muehlebach, Michael},
  booktitle = {ICML 2024 Workshops: RLControlTheory},
  year      = {2024},
  url       = {https://mlanthology.org/icmlw/2024/er2024icmlw-eventbased/}
}