FedVar: Federated Learning Algorithm with Weight Variation in Clients

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

5 Scopus citations

Abstract

Among the studies to solve statistical-related problems in federated learning (FL), there are many studies to improve the non-independent and identically distributed (Non-IID) problem of user data. This problem occurs because each local device collects data from different clients, so the size and characteristics of the data are different. When the data distribution of local devices is IID, it does not negatively affect learning, but when non-IID, it does not reach the general cloud computing performance. In this paper, we propose FedVar, an algorithm that uses the weight standard deviation for each client, to improve the situation in which the distribution of data is non-IID, so FL learning is not done properly. In addition, the performance is verified by comparing the acuity and loss of the proposed algorithm with other existing FL algorithms including FedAvg.

Original languageEnglish
Title of host publicationITC-CSCC 2022 - 37th International Technical Conference on Circuits/Systems, Computers and Communications
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages456-459
Number of pages4
ISBN (Electronic)9781665485593
DOIs
StatePublished - 2022
Externally publishedYes
Event37th International Technical Conference on Circuits/Systems, Computers and Communications, ITC-CSCC 2022 - Phuket, Thailand
Duration: 5 Jul 20228 Jul 2022

Publication series

NameITC-CSCC 2022 - 37th International Technical Conference on Circuits/Systems, Computers and Communications

Conference

Conference37th International Technical Conference on Circuits/Systems, Computers and Communications, ITC-CSCC 2022
Country/TerritoryThailand
CityPhuket
Period5/07/228/07/22

Keywords

  • Data Heterogeneity
  • FedAvg
  • FedSGD
  • FedVar
  • Federated Learning
  • Non-IID Data

Fingerprint

Dive into the research topics of 'FedVar: Federated Learning Algorithm with Weight Variation in Clients'. Together they form a unique fingerprint.

Cite this