FLUX: Efficient Descriptor-Driven Clustered Federated Learning under Arbitrary Distribution Shifts

TL;DR

FLUX introduces a clustering-based federated learning framework that adapts to arbitrary distribution shifts without prior knowledge, significantly improving accuracy and robustness in real-world scenarios.

Contribution

FLUX is a novel clustering-based federated learning method that handles unknown distribution shifts and supports test-time adaptation without prior shift information.

Findings

Achieves up to 23% accuracy gain over baselines.

Maintains computational and communication efficiency.

Effective across diverse benchmarks and real-world datasets.

Abstract

Federated Learning (FL) enables collaborative model training across multiple clients while preserving data privacy. Traditional FL methods often use a global model to fit all clients, assuming that clients' data are independent and identically distributed (IID). However, when this assumption does not hold, the global model accuracy may drop significantly, limiting FL applicability in real-world scenarios. To address this gap, we propose FLUX, a novel clustering-based FL (CFL) framework that addresses the four most common types of distribution shifts during both training and test time. To this end, FLUX leverages privacy-preserving client-side descriptor extraction and unsupervised clustering to ensure robust performance and scalability across varying levels and types of distribution shifts. Unlike existing CFL methods addressing non-IID client distribution shifts, FLUX i) does not require any prior knowledge of the types of distribution shifts or the number of client clusters, and ii) supports test-time adaptation, enabling unseen and unlabeled clients to benefit from the most suitable cluster-specific models. Extensive experiments across four standard benchmarks, two real-world datasets and ten state-of-the-art baselines show that FLUX improves performance and stability under diverse distribution shifts, achieving an average accuracy gain of up to 23 percentage points over the best-performing baselines, while maintaining computational and communication overhead comparable to FedAvg.