Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Graph Learning with Graphless Clients

Xingbo Fu, Song Wang, Yushun Dong, Binchi Zhang, Chen Chen, Jundong Li

TL;DR

In this paper, a novel framework FedGLS is proposed to tackle the problem in FGL with graphless clients, and extensive experiments demonstrate the superiority of the proposed FedGLS over five baselines.

Abstract

Federated Graph Learning (FGL) is tasked with training machine learning models, such as Graph Neural Networks (GNNs), for multiple clients, each with its own graph data. Existing methods usually assume that each client has both node features and graph structure of its graph data. In real-world scenarios, however, there exist federated systems where only a part of the clients have such data while other clients (i.e. graphless clients) may only have node features. This naturally leads to a novel problem in FGL: how to jointly train a model over distributed graph data with graphless clients? In this paper, we propose a novel framework FedGLS to tackle the problem in FGL with graphless clients. In FedGLS, we devise a local graph learner on each graphless client which learns the local graph structure with the structure knowledge transferred from other clients. To enable structure knowledge transfer, we design a GNN model and a feature encoder on each client. During local training, the feature encoder retains the local graph structure knowledge together with the GNN model via knowledge distillation, and the structure knowledge is transferred among clients in global update. Our extensive experiments demonstrate the superiority of the proposed FedGLS over five baselines.

Federated Graph Learning with Graphless Clients

TL;DR

In this paper, a novel framework FedGLS is proposed to tackle the problem in FGL with graphless clients, and extensive experiments demonstrate the superiority of the proposed FedGLS over five baselines.

Abstract

Federated Graph Learning (FGL) is tasked with training machine learning models, such as Graph Neural Networks (GNNs), for multiple clients, each with its own graph data. Existing methods usually assume that each client has both node features and graph structure of its graph data. In real-world scenarios, however, there exist federated systems where only a part of the clients have such data while other clients (i.e. graphless clients) may only have node features. This naturally leads to a novel problem in FGL: how to jointly train a model over distributed graph data with graphless clients? In this paper, we propose a novel framework FedGLS to tackle the problem in FGL with graphless clients. In FedGLS, we devise a local graph learner on each graphless client which learns the local graph structure with the structure knowledge transferred from other clients. To enable structure knowledge transfer, we design a GNN model and a feature encoder on each client. During local training, the feature encoder retains the local graph structure knowledge together with the GNN model via knowledge distillation, and the structure knowledge is transferred among clients in global update. Our extensive experiments demonstrate the superiority of the proposed FedGLS over five baselines.

Paper Structure

This paper contains 33 sections, 20 equations, 3 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Federated Learning
  4. Federated Graph Learning
  5. Problem Statement
  6. Preliminaries
  7. Problem Definition
  8. Methodology
  9. Framework Overview
  10. Local Training
  11. Global Update
  12. Overall Algorithm
  13. Experiments
  14. Experiment Setup
  15. Datasets
  16. ...and 18 more sections

Figures (3)

  • Figure 1: An example of a healthcare system including four hospitals. In this example, Hospital A and Hospital D have their local datasets of patients (node features) and co-staying information (links) among them. In the meantime, Hospital B and Hospital C only have their local datasets of patients (node features). The four hospitals aim to jointly train a model for predicting whether a patient is at high risk of contracting a contagious disease, orchestrated by a third-party company over their local datasets while the company cannot directly access their private datasets.
  • Figure 2: An overview of the proposed FedGLS.
  • Figure 3: Result for convergence speeds of FedGLS and Fed-GNNk: (a) training loss curve and (b) test accuracy curve on Cora; (c) training loss curve and (d) test accuracy curve on Flickr.