Provably Near-Optimal Federated Ensemble Distillation with Negligible Overhead

2025-02-10

AI Summary

In federated learning, high client heterogeneity severely degrades the quality of server-side pseudo-labels. Method: We propose a GAN-inspired discriminator-cooperation framework: lightweight discriminators are trained locally on clients, while generators are deployed distributively on the server; pseudo-label consistency is jointly optimized. We further introduce the first provably near-optimal client weighting mechanism to achieve high-quality pseudo-label aggregation. Integrating federated learning, knowledge distillation, and generative modeling, our approach improves global model performance without requiring server-side labeled data. Results: Our method consistently outperforms state-of-the-art baselines across multiple image classification benchmarks. It incurs negligible additional communication overhead (<0.5%), ensures controllable privacy leakage, and imposes minimal extra computational cost on clients (<3%).

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing

Abstract

Federated ensemble distillation addresses client heterogeneity by generating pseudo-labels for an unlabeled server dataset based on client predictions and training the server model using the pseudo-labeled dataset. The unlabeled server dataset can either be pre-existing or generated through a data-free approach. The effectiveness of this approach critically depends on the method of assigning weights to client predictions when creating pseudo-labels, especially in highly heterogeneous settings. Inspired by theoretical results from GANs, we propose a provably near-optimal weighting method that leverages client discriminators trained with a server-distributed generator and local datasets. Our experiments on various image classification tasks demonstrate that the proposed method significantly outperforms baselines. Furthermore, we show that the additional communication cost, client-side privacy leakage, and client-side computational overhead introduced by our method are negligible, both in scenarios with and without a pre-existing server dataset.

Problem

• Optimizes federated ensemble distillation for client heterogeneity
• Proposes near-optimal weighting method for pseudo-label generation
• Minimizes communication cost and client-side overhead in federated learning

Innovation

• Federated ensemble distillation
• Provably near-optimal weighting
• Negligible overhead and leakage