Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedGTA: Topology-aware Averaging for Federated Graph Learning

Xunkai Li, Zhengyu Wu, Wentao Zhang, Yinlin Zhu, Rong-Hua Li, Guoren Wang

TL;DR

FedGTA addresses a core gap in Federated Graph Learning by introducing topology-aware optimization that couples topology-informed signals with personalized aggregation. It uses a k-step Non-param LP to produce topology-aware soft labels, computes a local smoothing confidence via entropy, and extracts mixed moments of neighbor features to drive data-driven client selection for aggregation. Empirical results on 12 real-world datasets, including ogbn-papers100M, show consistent improvements over strong FGL baselines in both transductive and inductive settings and demonstrate broad generalization across diverse GNN backbones. The framework achieves superior efficiency and scalability by decoupling topology-aware computations from heavy local-model training, making it applicable to large-scale graph learning in federated settings.

Abstract

Federated Graph Learning (FGL) is a distributed machine learning paradigm that enables collaborative training on large-scale subgraphs across multiple local systems. Existing FGL studies fall into two categories: (i) FGL Optimization, which improves multi-client training in existing machine learning models; (ii) FGL Model, which enhances performance with complex local models and multi-client interactions. However, most FGL optimization strategies are designed specifically for the computer vision domain and ignore graph structure, presenting dissatisfied performance and slow convergence. Meanwhile, complex local model architectures in FGL Models studies lack scalability for handling large-scale subgraphs and have deployment limitations. To address these issues, we propose Federated Graph Topology-aware Aggregation (FedGTA), a personalized optimization strategy that optimizes through topology-aware local smoothing confidence and mixed neighbor features. During experiments, we deploy FedGTA in 12 multi-scale real-world datasets with the Louvain and Metis split. This allows us to evaluate the performance and robustness of FedGTA across a range of scenarios. Extensive experiments demonstrate that FedGTA achieves state-of-the-art performance while exhibiting high scalability and efficiency. The experiment includes ogbn-papers100M, the most representative large-scale graph database so that we can verify the applicability of our method to large-scale graph learning. To the best of our knowledge, our study is the first to bridge large-scale graph learning with FGL using this optimization strategy, contributing to the development of efficient and scalable FGL methods.

FedGTA: Topology-aware Averaging for Federated Graph Learning

TL;DR

FedGTA addresses a core gap in Federated Graph Learning by introducing topology-aware optimization that couples topology-informed signals with personalized aggregation. It uses a k-step Non-param LP to produce topology-aware soft labels, computes a local smoothing confidence via entropy, and extracts mixed moments of neighbor features to drive data-driven client selection for aggregation. Empirical results on 12 real-world datasets, including ogbn-papers100M, show consistent improvements over strong FGL baselines in both transductive and inductive settings and demonstrate broad generalization across diverse GNN backbones. The framework achieves superior efficiency and scalability by decoupling topology-aware computations from heavy local-model training, making it applicable to large-scale graph learning in federated settings.

Abstract

Federated Graph Learning (FGL) is a distributed machine learning paradigm that enables collaborative training on large-scale subgraphs across multiple local systems. Existing FGL studies fall into two categories: (i) FGL Optimization, which improves multi-client training in existing machine learning models; (ii) FGL Model, which enhances performance with complex local models and multi-client interactions. However, most FGL optimization strategies are designed specifically for the computer vision domain and ignore graph structure, presenting dissatisfied performance and slow convergence. Meanwhile, complex local model architectures in FGL Models studies lack scalability for handling large-scale subgraphs and have deployment limitations. To address these issues, we propose Federated Graph Topology-aware Aggregation (FedGTA), a personalized optimization strategy that optimizes through topology-aware local smoothing confidence and mixed neighbor features. During experiments, we deploy FedGTA in 12 multi-scale real-world datasets with the Louvain and Metis split. This allows us to evaluate the performance and robustness of FedGTA across a range of scenarios. Extensive experiments demonstrate that FedGTA achieves state-of-the-art performance while exhibiting high scalability and efficiency. The experiment includes ogbn-papers100M, the most representative large-scale graph database so that we can verify the applicability of our method to large-scale graph learning. To the best of our knowledge, our study is the first to bridge large-scale graph learning with FGL using this optimization strategy, contributing to the development of efficient and scalable FGL methods.
Paper Structure (16 sections, 7 equations, 6 figures, 6 tables, 2 algorithms)

This paper contains 16 sections, 7 equations, 6 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. PRELIMINARIES AND RELATED WORKS
  3. Problem Formulation
  4. Conventional and Scalable GNNs
  5. FGL Optimization
  6. FGL Model
  7. FedGTA FRAMEWORK
  8. The Proposed FedGTA
  9. Complexity Analysis
  10. Experiments
  11. Experimental Setup
  12. Performance Comparison
  13. Generalization
  14. Method Interpretability
  15. Efficiency and Scalability Analysis
  16. ...and 1 more sections

Figures (6)

  • Figure 1: Empirical analysis on Cora in 10 clients with GCN, which contains 7 different node labels. (a) The color from white to blue represents the number of nodes held by different clients in each class a gradual increase in quantity. (b) The x-axis of the line plot is the federated training round. "Global" and "Local" represent the model performance in centralized and siloed settings, respectively.
  • Figure 2: Overview of our proposed FedGTA framework. The different colors of the nodes represent the different labels.
  • Figure 3: The visualization of model aggregation in Amazon Photo with 10 clients split, which contains 8 different node labels. The circle size corresponds to aggregation weight, from large to small.
  • Figure 4: Convergence curves of our proposed FedGTA and baseline models on $4$ large-scale benchmark graph datasets. Curves represent the local model training phase and model aggregation phase on the server side. The shaded area is the result range of multiple runs.
  • Figure 5: Training efficiency with different numbers of clients.
  • ...and 1 more figures