Empowering Trustworthy Client Selection in Edge Federated Learning Leveraging Reinforcement Learning

Asadullah Tariq, Abderrahmane Lakas, Farag M. Sallabi, Tariq Qayyum, Mohamed Serhani, Ezedin Baraka

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Federated learning (FL) is a promising approach for training AI models across multiple clients in Edge Computing (EC), without sharing raw local data. By enabling local training and aggregating updates into a global model, FL maintains privacy while facilitating collaborative learning. Nevertheless, FL encounters several challenges, including trustworthy client participation, inefficient model aggregation due to client with malicious or less accurate model. In this paper, we propose a trustworthy FL method incorporating Q-learning, trust, and reputation mechanisms, enhancing model accuracy and fairness. This method promotes client participation, mitigates malicious attacks' impact, and ensures fair model distribution. Inspired by reinforcement learning, the Q-learning algorithm optimizes client selection using the Bellman equation, enabling the server to balance exploration and exploitation for improved system performance. Furthermore, we explored the advantages of peer-to-peer FL settings. Extensive experimentation demonstrates our proposed trustworthy FL approach's effectiveness in achieving high learning accuracy while ensuring fairness across clients and maintaining efficient client selection. Our results reveal significant improvements in model performance, convergence speed, and generalization.

Original languageEnglish
Title of host publicationProceedings - 2023 IEEE/ACM Symposium on Edge Computing, SEC 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages372-377
Number of pages6
ISBN (Electronic)9798400701238
DOIs
Publication statusPublished - 2023
Event8th Annual IEEE/ACM Symposium on Edge Computing, SEC 2023 - Wilmington, United States
Duration: Dec 6 2023Dec 9 2023

Publication series

NameProceedings - 2023 IEEE/ACM Symposium on Edge Computing, SEC 2023

Conference

Conference8th Annual IEEE/ACM Symposium on Edge Computing, SEC 2023
Country/TerritoryUnited States
CityWilmington
Period12/6/2312/9/23

Keywords

  • Edge Computing
  • Federated Learning
  • Privacy
  • Reinforcement learning
  • Reputation
  • Trust

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Networks and Communications
  • Hardware and Architecture
  • Computer Science Applications

Fingerprint

Dive into the research topics of 'Empowering Trustworthy Client Selection in Edge Federated Learning Leveraging Reinforcement Learning'. Together they form a unique fingerprint.

Cite this