Novel clustered federated learning based on local loss

TL;DR

This paper introduces LCFL, a new clustering metric for federated learning that accurately assesses client data distribution variations without prior knowledge, enhancing privacy and performance.

Contribution

The paper presents LCFL, a novel clustering metric tailored for federated learning, addressing privacy, non-convex models, and accuracy issues in existing methods.

Findings

LCFL outperforms baseline methods on multiple benchmarks.

LCFL accurately evaluates client data distribution variations.

The framework is mathematically rigorous and feasible.

Abstract

This paper proposes LCFL, a novel clustering metric for evaluating clients' data distributions in federated learning. LCFL aligns with federated learning requirements, accurately assessing client-to-client variations in data distribution. It offers advantages over existing clustered federated learning methods, addressing privacy concerns, improving applicability to non-convex models, and providing more accurate classification results. LCFL does not require prior knowledge of clients' data distributions. We provide a rigorous mathematical analysis, demonstrating the correctness and feasibility of our framework. Numerical experiments with neural network instances highlight the superior performance of LCFL over baselines on several clustered federated learning benchmarks.