Personalized Federated Learning With Similarity Information Supervisor

27 Sept 2024 (modified: 05 Feb 2025)Submitted to ICLR 2025EveryoneRevisionsBibTeXCC BY 4.0
Keywords: Personalized Federated Learning, Heterogeneous Data
Abstract:

A crucial issue in federated learning is the heterogeneity of data between clients, which can lead to model weight divergence, eventually deteriorating the model performance. Personalized federated learning (pFL) has been proven to be an effective approach to addressing data heterogeneity in federated learning. However, existing pFL studies seldom verify whether the broadcast global model is beneficial for the local model performance. To address this, we propose a novel pFL method, called federated learning with similarity information supervision (FedSimSup). Specifically, FedSimSup incorporates a local supervisor to assist the model training and a personalized model for global information aggregation. The role of the supervisor is to refine the personalized model when it is not beneficial for the local model performance, ensuring the effective global information aggregation while aligning with the local heterogeneous data. Additionally, the similarity relationships between the clients are measured using label distribution differences of the local raw data to weight the personalized models, promoting information usage among similar clients. Experimental results demonstrate three advantages of FedSimSup: (1) It shows better performance over heterogeneous data compared with seven state-of-the-art federated learning methods; (2) It can allow for different model architectures across different clients; (3) It offers a certain degree of interpretability.

Primary Area: transfer learning, meta learning, and lifelong learning
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: 9285
Loading

OpenReview is a long-term project to advance science through improved peer review with legal nonprofit status. We gratefully acknowledge the support of the OpenReview Sponsors. © 2025 OpenReview