Heterogeneity-Aware Knowledge Sharing for Graph Federated Learning (ICML 2026)

Keywords: Graph Federated Learning, Heterogeneity, Dual-aspect Knowledge Sharing, Semantic Knowledge, Structural Knowledge

Abstract: Graph Federated Learning (GFL) enables distributed graph representation learning while protecting the privacy of graph data. However, GFL suffers from heterogeneity arising from diverse node features and structural topologies across multiple clients. To address both types of heterogeneity, we propose a novel graph Federated learning method via Semantic and Structural Alignment (FedSSA), which shares the knowledge of both node features and structural topologies among clients. For node feature heterogeneity, we propose a novel variational model to infer class-wise node distributions, so that we can cluster clients based on inferred distributions and construct cluster-level representative distributions. Then, we minimize the divergence between local and cluster-level distributions to facilitate semantic knowledge sharing. For structure heterogeneity, we employ spectral Graph Neural Networks (GNNs) and propose a novel spectral energy measure to characterize structural information, so that we can cluster clients based on spectral energy and build cluster-level spectral GNNs. Then, we align the spectral characteristics of local spectral GNNs with those of cluster-level spectral GNNs to enable structural knowledge sharing. Experiments on six homophilic and five heterophilic graph datasets under both non-overlapping and overlapping partitions demonstrate that FedSSA consistently outperforms eleven state-of-the-art methods.

The Table of Contents

• 😬 Dependencies and Installation
• 🥳 How to Run
• 🌹 Experimental Results
• 📕 License
• ☎️ Contact

😉 If FedSSA is helpful to you, please star this repo. Thanks! 🤗

😬 Dependencies and Installation

Before running or modifying the code, you need to:

• Make sure Anaconda or Miniconda is installed.

• Clone this repo to your machine.

  # git clone this repository
  git clone https://github.com/blgpb/FedSSA
  cd FedSSA
  
  # create a new Anaconda env 
  conda create -n fedssa python=3.8 -y
  conda activate fedssa  

• Install required packages

  # install python dependencies
  pip install -r requirements.txt

• It is recommended to run experiments via NVIDIA GeForce RTX 4090!

Requirements:

• Python 3.8.8
• PyTorch 1.12.0+cu113
• PyTorch Geometric 2.5.1
• NumPy, SciPy, and other dependencies listed in requirements.txt

🥳 How to Run

Basic Usage

python main.py --dataset Cora --n_clients 20 --mode disjoint

Parameter Description

• --dataset: Dataset name (Cora, CiteSeer, PubMed, Computers, Photo, ogbn-arxiv, Roman-empire, Amazon-ratings, Minesweeper, Tolokers, Questions)
• --mode: Partition mode (disjoint for non-overlapping, overlapping for overlapping partitions)
• --n-clients: Number of clients (default: 10)

Dataset Preparation

Download Pre-processed Datasets

Download from the Google Drive (https://drive.google.com/file/d/1PyqvR6yL43Om42fdsbKHj5WCgREvi3St/view?usp=sharing) and then unzip it. Place the datasets folder in the same path as README.md.

🌹 Experimental Results

FedSSA demonstrates superior performance across diverse graph datasets:

• Homophilic Datasets: Consistent improvements across Cora, CiteSeer, PubMed, Amazon-Computer, Amazon-Photo, and ogbn-arxiv
• Heterophilic Datasets: Average improvement of 5.79% over the second-best method on all heterophilic datasets (Roman-empire, Amazon-ratings, Minesweeper, Tolokers, Questions)
• Robustness: Superior performance under both non-overlapping and overlapping partition settings

Key Features

• Dual-Aspect Knowledge Sharing: Addresses both node feature heterogeneity and structural topology heterogeneity
• Semantic Knowledge Sharing: Variational model-based inference of class-wise node distributions with cluster-level alignment
• Structural Knowledge Sharing: Spectral GNN with novel spectral energy measure for structural information capture and alignment
• Scalability: Efficient distributed computation across multiple clients
• Versatility: Comprehensive coverage of 11 graph datasets with both homophilic and heterophilic properties

📕 License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0). See LICENSE.txt for details.

☎️ Contact

If you have any questions or suggestions, please feel free to contact us.

Star History

If you find this work helpful, please consider giving us a star! ⭐