Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BoostFGL: Boosting Fairness in Federated Graph Learning

Zekai Chen, Kairui Yang, Xunkai Li, Henan Sun, Zhihan Zhang, Jia Li, Qiangqiang Dai, Rong-Hua Li, Guoren Wang

TL;DR

BoostFGL tackles node-level fairness in federated graph learning by diagnosing three coupled sources of disparity—label skew, topology confounding, and aggregation dilution—and proposing a modular boosting framework with three mechanisms: client-side node boosting, client-side topology boosting, and server-side model boosting. Each component targets a specific stage of the client–server pipeline, and together they yield monotone improvements in process-level signals such as Gradient Share Disparity, Edge-wise Propagation Reliability, and Dilution Ratio, while remaining compatible with standard backbones and privacy mechanisms. Theoretical guarantees and asymptotic consistency show that BoostFGL reduces to standard FedAvg in high-confidence regimes. Empirically, BoostFGL achieves substantial fairness gains (e.g., Overall-F1 improvement of 8.43% across nine datasets) with competitive overall performance and good scalability, including large graphs where rivals fail due to memory limits. The results advocate a practical design principle: diagnose unfairness with stage-specific signals and apply coordinated, lightweight corrections at the responsible stage to realize robust, pipeline-aware fairness in federated graph learning.

Abstract

Federated graph learning (FGL) enables collaborative training of graph neural networks (GNNs) across decentralized subgraphs without exposing raw data. While existing FGL methods often achieve high overall accuracy, we show that this average performance can conceal severe degradation on disadvantaged node groups. From a fairness perspective, these disparities arise systematically from three coupled sources: label skew toward majority patterns, topology confounding in message propagation, and aggregation dilution of updates from hard clients. To address this, we propose \textbf{BoostFGL}, a boosting-style framework for fairness-aware FGL. BoostFGL introduces three coordinated mechanisms: \ding{182} \emph{Client-side node boosting}, which reshapes local training signals to emphasize systematically under-served nodes; \ding{183} \emph{Client-side topology boosting}, which reallocates propagation emphasis toward reliable yet underused structures and attenuates misleading neighborhoods; and \ding{184} \emph{Server-side model boosting}, which performs difficulty- and reliability-aware aggregation to preserve informative updates from hard clients while stabilizing the global model. Extensive experiments on 9 datasets show that BoostFGL delivers substantial fairness gains, improving Overall-F1 by 8.43\%, while preserving competitive overall performance against strong FGL baselines.

BoostFGL: Boosting Fairness in Federated Graph Learning

TL;DR

BoostFGL tackles node-level fairness in federated graph learning by diagnosing three coupled sources of disparity—label skew, topology confounding, and aggregation dilution—and proposing a modular boosting framework with three mechanisms: client-side node boosting, client-side topology boosting, and server-side model boosting. Each component targets a specific stage of the client–server pipeline, and together they yield monotone improvements in process-level signals such as Gradient Share Disparity, Edge-wise Propagation Reliability, and Dilution Ratio, while remaining compatible with standard backbones and privacy mechanisms. Theoretical guarantees and asymptotic consistency show that BoostFGL reduces to standard FedAvg in high-confidence regimes. Empirically, BoostFGL achieves substantial fairness gains (e.g., Overall-F1 improvement of 8.43% across nine datasets) with competitive overall performance and good scalability, including large graphs where rivals fail due to memory limits. The results advocate a practical design principle: diagnose unfairness with stage-specific signals and apply coordinated, lightweight corrections at the responsible stage to realize robust, pipeline-aware fairness in federated graph learning.

Abstract

Federated graph learning (FGL) enables collaborative training of graph neural networks (GNNs) across decentralized subgraphs without exposing raw data. While existing FGL methods often achieve high overall accuracy, we show that this average performance can conceal severe degradation on disadvantaged node groups. From a fairness perspective, these disparities arise systematically from three coupled sources: label skew toward majority patterns, topology confounding in message propagation, and aggregation dilution of updates from hard clients. To address this, we propose \textbf{BoostFGL}, a boosting-style framework for fairness-aware FGL. BoostFGL introduces three coordinated mechanisms: \ding{182} \emph{Client-side node boosting}, which reshapes local training signals to emphasize systematically under-served nodes; \ding{183} \emph{Client-side topology boosting}, which reallocates propagation emphasis toward reliable yet underused structures and attenuates misleading neighborhoods; and \ding{184} \emph{Server-side model boosting}, which performs difficulty- and reliability-aware aggregation to preserve informative updates from hard clients while stabilizing the global model. Extensive experiments on 9 datasets show that BoostFGL delivers substantial fairness gains, improving Overall-F1 by 8.43\%, while preserving competitive overall performance against strong FGL baselines.
Paper Structure (77 sections, 4 theorems, 42 equations, 15 figures, 4 tables, 4 algorithms)

This paper contains 77 sections, 4 theorems, 42 equations, 15 figures, 4 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem Formulation
  4. Related Work
  5. Graph neural networks.
  6. Federated graph learning.
  7. Fairness in federated learning.
  8. Fairness-aware federated graph learning.
  9. Positioning of BoostFGL.
  10. Empirical Study & Theoretical Analysis
  11. Node-side label skew
  12. Topology Confounding in Message Passing
  13. Server-side Aggregation Dilution
  14. Theoretical Analysis
  15. Node-side boosting increases gradient-share parity (GSD).
  16. ...and 62 more sections

Key Result

Lemma 3.1

Let $\mathrm{GSD}_{\textsc{base}}^{(t)}$ be computed under uniform weighting ($\alpha_v^{(t)} \equiv 1$), and $\mathrm{GSD}_{\textsc{boost}}^{(t)}$ under BoostFGL node weights. Under mild difficulty--gradient coupling and $\mathbb{E}[\bar{d} \mid \min] > \mathbb{E}[\bar{d} \mid \mathrm{maj}]$,

Figures (15)

  • Figure 1: Label skew diagnosis. Gradient Share Disparity (GSD) over federated rounds. The dashed line denotes the fair allocation baseline (GSD$=1$).
  • Figure 2: Topology confounding diagnosis. Distribution of Edge-wise Propagation Reliability (EPR). Negative EPR indicates harmful message passing.
  • Figure 3: K-hop error amplification. Error propagation as a function of distance from minority/heterophilous nodes. BoostFGL suppresses amplification under message passing.
  • Figure 4: Overview of the proposed BoostFGL framework. Each client applies node-side and topology-side boosting to amplify hard nodes and critical edges before local training, then uploads the updated model and statistics. The server performs model-side, layer-wise boosting during aggregation to obtain the global model for the next communication round.
  • Figure 5: Hyperparameter sensitivity.Left: Relative F1 drop under different node-boosting strengths $\lambda$ on representative datasets. Right: Heatmap over topology- and model-boosting hyperparameters $(\tau,\text{edge budget})$.
  • ...and 10 more figures

Theorems & Definitions (8)

  • Lemma 3.1: Gradient-share rectification
  • Theorem 3.2: Harmful-message suppression
  • Theorem 3.3: Trust-gated aggregation improves DR and bounds influence
  • Proposition 3.4: Reduction to standard FedAvg
  • proof : Proof of Lemma \ref{['lem:gsd_rect']}
  • proof : Proof of Theorem \ref{['thm:epr_suppress']}
  • proof : Proof of Theorem \ref{['thm:dr_improve']}
  • proof : Proof of Proposition \ref{['prop:consistency']}