Submit

Login Login

Register Register

Home / Journals / CMC / Online First / doi:10.32604/cmc.2026.086553
Journal Menu
Special Issues
Table of Content

All issues

Online First

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Open Access

ARTICLE

FedCANA: Communication-Efficient Adaptive Neighborhood Aggregation for Personalized Federated Learning under Non-IID Data

Yao Pu, Rou Zhou, Yuling Chen*
State Key Laboratory of Public Big Data, College of Computer Science and Technology, Guizhou University, Guiyang, China
* Corresponding Author: Yuling Chen. Email: email

Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.086553

Received 01 June 2026; Accepted 26 August 2026; Published online 14 September 2026

Abstract

Personalized federated learning (PFL) aims to address the insufficient adaptability of a single global model caused by Non-IID data. It improves client-specific performance by learning customized models for different clients. However, existing PFL methods based on similarity modeling often struggle to balance personalization performance and communication efficiency under Non-IID settings. To address this problem, this paper proposes FedCANA (Communication-Efficient Adaptive Neighborhood Aggregation), a communication-efficient personalized federated learning method for Non-IID data. First, FedCANA adopts Residual-Spatial Attention Collaborative (RSAC) module to strengthen the clients’ local representation ability, which provides a more stable model basis for subsequent similarity estimation. Second, FedCANA employs Saliency-Guided Differential Descriptor (SGDD) to select and upload critical residual information, which reduces redundant communication while preserving client-specific collaborative signals. Finally, FedCANA introduces Entropy-driven Adaptive Neighborhood Aggregation (EANA) to dynamically select similar clients for collaborative aggregation, which reduces the interference from less relevant clients and improves personalized model performance. Extensive experiments on three standard benchmark datasets under three Non-IID settings demonstrate that FedCANA achieves state-of-the-art personalized accuracy on most tasks, and yields up to 9.26% accuracy improvement on the highly heterogeneous CIFAR-100 dataset. Meanwhile, it maintains significantly lower communication overhead than mainstream similarity-based personalized federated learning methods, achieving a favorable trade-off between accuracy and communication efficiency.

Keywords

Personalized federated learning; data heterogeneity; communication efficiency; machine learning; adaptive aggregation

  • 69

    View

  • 16

    Download

  • 0

    Like

Related articles
Copyright© 2026 Tech Science Press
© 1997-2026 TSP (Henderson, USA) unless otherwise stated

Share Link