One-Shot Federated Learning-based Model-Free Reinforcement Learning

Gaith Rjoub, Jamal Bentahar, Omar Abdel Wahab, Nagat Drawel

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The Federated Learning (FL) paradigm is emerging as a way to train machine learning (ML) models in distributed systems. A large population of interconnected devices (i.e. Internet of Things (IoT)) acting as local learners optimize the model parameters collectively (e.g., neural networks’ weights), rather than sharing and disclosing the training data set with the server. FL approaches assume each participant has enough training data for the tasks of interest. Realistically, data collected by IoT devices may be insufficient and often unlabeled. In particular, each IoT device may only contain one or a few samples of every relevant data category, and may not have the time or interest to label them. In realistic applications, this severely limits FL’s practicality and usability. In this paper, we propose a One-Shot Federated Learning (OSFL) framework considering a FL scenario wherein the local training is carried out on IoT devices and the global aggregation is done at the level of an edge server. Moreover, we combine model-free reinforcement learning with OSFL to design a more intelligent IoT device to infer whether to label a sample automatically or request the true label for the one-shot learning set-up. We validate our system on the SODA10M dataset. Experiments show that our solution achieves better performance than DQN and RS benchmark approaches.

Original languageBritish English
Title of host publicationThe International Conference on Deep Learning, Big Data and Blockchain, DBB 2022
EditorsIrfan Awan, Muhammad Younas, Jamal Bentahar, Salima Benbernou
PublisherSpringer Science and Business Media Deutschland GmbH
Pages39-52
Number of pages14
ISBN (Print)9783031160349
DOIs
StatePublished - 2023
Event3rd International Conference on Deep Learning, Big Data and Blockchain, DBB 2022 - Rome, Italy
Duration: 22 Aug 202224 Aug 2022

Publication series

NameLecture Notes in Networks and Systems
Volume541 LNNS
ISSN (Print)2367-3370
ISSN (Electronic)2367-3389

Conference

Conference3rd International Conference on Deep Learning, Big Data and Blockchain, DBB 2022
Country/TerritoryItaly
CityRome
Period22/08/2224/08/22

Keywords

  • Federated learning
  • Internet of things
  • Model-free reinforcement learning
  • One-shot learning

Fingerprint

Dive into the research topics of 'One-Shot Federated Learning-based Model-Free Reinforcement Learning'. Together they form a unique fingerprint.

Cite this