Interaction Based Gaussian Weighting Clustering for Federated Learning

Alessandro Licciardi, Davide Leo, Eros Fanì, Barbara Caputo, Marco Ciccone

Keywords: Federated Learning, Clustered Federated Learning, Personalized Federated Learning

TL;DR: We present a novel approach for clustering clients in FL to mitigate the effects of heterogeneity and class-imbalance within clients' data distribution. We motivate our work with a comprehensive mathematical framework

Abstract: Federated learning emerged as a decentralized paradigm to train models while securing privacy. However, conventional FL faces data heterogeneity and class imbalance challenges, affecting model performance. In response to these issues, Personalized FL has been developed as an innovative methodology that relies on fine-tuning the distinct local models based on individual training datasets. In this work, we propose a novel PFL method, FedGW (Federated Gaussian Weighting), which groups clients based on their data distribution, allowing training of a more robust and personalized model on the identified clusters. FedGW identifies homogeneous clusters by transforming individual empirical losses to model client interactions with a Gaussian reward mechanism. Additionally, we introduce a new clustering metric for FL to evaluate cluster cohesion with respect to the individual class distribution. Our experiments on benchmark datasets show that FedGW outperforms existing FL algorithms in cluster quality and classification accuracy, validating the efficacy of our approach.

Primary Area: other topics in machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9633