Resource-Efficient Personalization in Federated Learning With Closed-Form Classifiers

Statistical heterogeneity in Federated Learning (FL) often leads to client drift and biased local solutions. Prior work in the literature shows that client drift particularly affects the parameters of the classification layer, hindering both convergence and accuracy. While Personalized FL (PFL) addr...

Full description

Saved in:
Bibliographic Details
Main Authors: Eros Fani, Raffaello Camoriano, Barbara Caputo, Marco Ciccone
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10946159/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849731639364550656
author Eros Fani
Raffaello Camoriano
Barbara Caputo
Marco Ciccone
author_facet Eros Fani
Raffaello Camoriano
Barbara Caputo
Marco Ciccone
author_sort Eros Fani
collection DOAJ
description Statistical heterogeneity in Federated Learning (FL) often leads to client drift and biased local solutions. Prior work in the literature shows that client drift particularly affects the parameters of the classification layer, hindering both convergence and accuracy. While Personalized FL (PFL) addresses this by allowing client-specific models, it can overlook valuable global knowledge. This paper introduces Federated Recursive Ridge Regression (<monospace>Fed3R</monospace>), a fast and efficient method to construct a closed-form classifier that effectively incorporates global knowledge while being inherently robust to statistical heterogeneity. <monospace>Fed3R</monospace> leverages a pre-trained feature extractor and a recursive ridge regression formulation to achieve exact aggregation of local classifiers and recover the centralized solution. We demonstrate that <monospace>Fed3R</monospace> serves as a robust initialization for further fine-tuning with various FL and PFL algorithms, accelerating convergence and boosting performance. Furthermore, we propose Only Local Labels (<monospace>OLL</monospace>), a novel PFL technique that simplifies local classifiers by focusing only on locally relevant classes, preventing misclassifications and improving efficiency. Our empirical evaluation on real-world cross-device datasets shows that <monospace>Fed3R</monospace>, combined with <monospace>OLL</monospace>, significantly improves performance and reduces training costs in heterogeneous FL and PFL scenarios.
format Article
id doaj-art-59f7ea047d5444168ca93dbf1204e6c4
institution DOAJ
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-59f7ea047d5444168ca93dbf1204e6c42025-08-20T03:08:28ZengIEEEIEEE Access2169-35362025-01-0113619286195710.1109/ACCESS.2025.355658710946159Resource-Efficient Personalization in Federated Learning With Closed-Form ClassifiersEros Fani0https://orcid.org/0000-0002-4486-1086Raffaello Camoriano1https://orcid.org/0000-0002-8890-2732Barbara Caputo2https://orcid.org/0000-0001-7169-0158Marco Ciccone3https://orcid.org/0000-0002-3306-1323Department of Control and Computer Engineering, Polytechnic University of Turin, Turin, ItalyDepartment of Control and Computer Engineering, Polytechnic University of Turin, Turin, ItalyDepartment of Control and Computer Engineering, Polytechnic University of Turin, Turin, ItalyVector Institute, Toronto, ON, CanadaStatistical heterogeneity in Federated Learning (FL) often leads to client drift and biased local solutions. Prior work in the literature shows that client drift particularly affects the parameters of the classification layer, hindering both convergence and accuracy. While Personalized FL (PFL) addresses this by allowing client-specific models, it can overlook valuable global knowledge. This paper introduces Federated Recursive Ridge Regression (<monospace>Fed3R</monospace>), a fast and efficient method to construct a closed-form classifier that effectively incorporates global knowledge while being inherently robust to statistical heterogeneity. <monospace>Fed3R</monospace> leverages a pre-trained feature extractor and a recursive ridge regression formulation to achieve exact aggregation of local classifiers and recover the centralized solution. We demonstrate that <monospace>Fed3R</monospace> serves as a robust initialization for further fine-tuning with various FL and PFL algorithms, accelerating convergence and boosting performance. Furthermore, we propose Only Local Labels (<monospace>OLL</monospace>), a novel PFL technique that simplifies local classifiers by focusing only on locally relevant classes, preventing misclassifications and improving efficiency. Our empirical evaluation on real-world cross-device datasets shows that <monospace>Fed3R</monospace>, combined with <monospace>OLL</monospace>, significantly improves performance and reduces training costs in heterogeneous FL and PFL scenarios.https://ieeexplore.ieee.org/document/10946159/Federated learningpersonalized federated learningpre-trained modelsridge regressionclosed-form classifiers
spellingShingle Eros Fani
Raffaello Camoriano
Barbara Caputo
Marco Ciccone
Resource-Efficient Personalization in Federated Learning With Closed-Form Classifiers
IEEE Access
Federated learning
personalized federated learning
pre-trained models
ridge regression
closed-form classifiers
title Resource-Efficient Personalization in Federated Learning With Closed-Form Classifiers
title_full Resource-Efficient Personalization in Federated Learning With Closed-Form Classifiers
title_fullStr Resource-Efficient Personalization in Federated Learning With Closed-Form Classifiers
title_full_unstemmed Resource-Efficient Personalization in Federated Learning With Closed-Form Classifiers
title_short Resource-Efficient Personalization in Federated Learning With Closed-Form Classifiers
title_sort resource efficient personalization in federated learning with closed form classifiers
topic Federated learning
personalized federated learning
pre-trained models
ridge regression
closed-form classifiers
url https://ieeexplore.ieee.org/document/10946159/
work_keys_str_mv AT erosfani resourceefficientpersonalizationinfederatedlearningwithclosedformclassifiers
AT raffaellocamoriano resourceefficientpersonalizationinfederatedlearningwithclosedformclassifiers
AT barbaracaputo resourceefficientpersonalizationinfederatedlearningwithclosedformclassifiers
AT marcociccone resourceefficientpersonalizationinfederatedlearningwithclosedformclassifiers