DFED: Data-Free Ensemble Distillation with Multi-Source GANs for Heterogeneous Federated Learning
Keywords: Federated learning, Data heterogeneity, Data-Free Knowledge Distillation
Abstract: Federated Learning (FL) is a decentralized machine learning paradigm that enables clients to collaboratively train models while preserving data privacy. However, surmounting the obstacles introduced by data heterogeneity in heterogeneous federated learning remains a profound challenge, as it drives each client towards distinct convergence trajectories, impeding the global model's convergence. To transcend these challenges, we propose DFED, a novel data-free ensemble knowledge distillation method designed to counteract the effects of data heterogeneity. DFED leverages multi-source Generative Adversarial Networks (GANs) to generate synthetic data that aligns with local distributions, ensuring privacy while promoting diverse feature representations across clients. Additionally, DFED aggregates client models into an ensemble based on their specialized knowledge, and applies ensemble distillation to refine the global model, mitigating the issues caused by disparities in data distributions. Across a variety of image classification benchmarks, DFED demonstrates superior performance compared to several state-of-the-art (SOTA) methods. The source code will be made publicly accessible once the paper has been accepted for publication.
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.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 3489
Loading