FedHarmony: Harmonizing Heterogeneous Label Correlations in Federated Multi-Label Learning

TL;DR

FedHarmony is a novel framework that aligns heterogeneous label correlations in federated multi-label learning, improving global model accuracy while preserving privacy.

Contribution

It introduces a consensus correlation mechanism and an accelerated optimization algorithm, addressing label correlation drift in federated multi-label learning.

Findings

FedHarmony outperforms existing methods on real-world datasets.

The framework achieves faster convergence without accuracy loss.

It effectively harmonizes heterogeneous label correlations across clients.

Abstract

Federated Multi-Label Learning is a distributed paradigm where multiple clients possess heterogeneous multi-label data and perform collaborative learning under privacy constraints without sharing raw data. However, modeling label correlations under heterogeneous distributions remains challenging. Due to client-specific label spaces and varying co-occurrence patterns, correlations learned by individual clients inevitably deviate from the global structure, a phenomenon we term label correlation drift. To address this, we propose FedHarmony, a framework that harmonizes heterogeneous label correlations across clients. It introduces consensus correlation, capturing agreement among other clients and serving as a global teacher to correct biased local estimates. During aggregation, FedHarmony evaluates each client by both data size and correlation quality, assigning weights accordingly. Moreover, we develop an accelerated optimization algorithm for FedHarmony and theoretically establish faster convergence without sacrificing accuracy. Experiments on real-world federated multi-label datasets show that FedHarmony consistently outperforms state-of-the-art methods.