Towards Unbiased Federated Graph Learning: Label and Topology Perspectives

TL;DR

This work tackles fairness in Federated Graph Learning under data and topology heterogeneity. It introduces FairFGL, a fairness-centered framework with History-Preserving, Majority Alignment, and Gradient Modification on the client side, plus Top-$k$ parameter sharing and clustering-based aggregation on the server to reflect local distributions and curb global majority bias. Empirical results across eight benchmarks show that FairFGL yields substantial improvements in minority-group Macro-F1 (up to 22.62% in some cases) while maintaining competitive overall accuracy and achieving faster convergence. The approach advances practical, fairness-aware subgraph-FL, enabling more equitable node representations in heterogenous decentralized graphs.

Abstract

Federated Graph Learning (FGL) enables privacy-preserving, distributed training of graph neural networks without sharing raw data. Among its approaches, subgraph-FL has become the dominant paradigm, with most work focused on improving overall node classification accuracy. However, these methods often overlook fairness due to the complexity of node features, labels, and graph structures. In particular, they perform poorly on nodes with disadvantaged properties, such as being in the minority class within subgraphs or having heterophilous connections (neighbors with dissimilar labels or misleading features). This reveals a critical issue: high accuracy can mask degraded performance on structurally or semantically marginalized nodes. To address this, we advocate for two fairness goals: (1) improving representation of minority class nodes for class-wise fairness and (2) mitigating topological bias from heterophilous connections for topology-aware fairness. We propose FairFGL, a novel framework that enhances fairness through fine-grained graph mining and collaborative learning. On the client side, the History-Preserving Module prevents overfitting to dominant local classes, while the Majority Alignment Module refines representations of heterophilous majority-class nodes. The Gradient Modification Module transfers minority-class knowledge from structurally favorable clients to improve fairness. On the server side, FairFGL uploads only the most influenced subset of parameters to reduce communication costs and better reflect local distributions. A cluster-based aggregation strategy reconciles conflicting updates and curbs global majority dominance . Extensive evaluations on eight benchmarks show FairFGL significantly improves minority-group performance , achieving up to a 22.62 percent Macro-F1 gain while enhancing convergence over state-of-the-art baselines.

Figures (5)

• Figure 1: Collaborative pandemic analysis across regions with edge representing social interactions between patients.
• Figure 2: Nodes volume of each class within each client is presented proportionally with indications of majority classes at the bottom. For fair comparison, we rank nodes by homophily score, selecting the top 50% as homophilous and the rest as heterophilous. The right-hand side showcases performances of four representative FGL methods: 1) Top-Side: Individual method performance on Cora under dual-perspective partitioning strategies; 2) Bottom-side: Comprehensive Evaluation concerning Heterophilous (Hete) nodes particularly.
• Figure 3: Describe the pipeline of FairFGL's training procedure. $t$ indicates the $t$-th training round. History-Preserving Module leverage the received History Model to regulate model training; Majority Alignment Module rectifies the Hete-Maj nodes through acquired class-wise prototype via KL loss function; Gradient Modification Module rectifies the gradients directionalities of local model.
• Figure 4: Hyperparameter Analysis on Cora Dataset.
• Figure 5: Convergency Test on Cora, Minesweeper, and PubMed datasets.