Graph Privacy: A Heterogeneous Federated GNN for Trans-Border Financial Data Circulation

TL;DR

This work tackles privacy-preserving trans-border financial data sharing by introducing HFGNN, a heterogeneous federated Graph Neural Network that trains local subgraphs on edge servers and uses a central server to securely manage aggregated information, preserving data confidentiality. The method explicitly separates and combines topological and feature information across heterogeneous subgraphs, enabling personalized local models while benefiting from shared knowledge. The authors demonstrate through simulations on EMNIST and CIFAR-10 that HFGNN outperforms traditional federated baselines and maintains robustness as the number of subgraphs grows, with faster convergence and improved accuracy. The approach has practical impact for regulated, cross-border financial data analysis by delivering joint modeling capabilities without exposing sensitive data.

Abstract

The sharing of external data has become a strong demand of financial institutions, but the privacy issue has led to the difficulty of interconnecting different platforms and the low degree of data openness. To effectively solve the privacy problem of financial data in trans-border flow and sharing, to ensure that the data is available but not visible, to realize the joint portrait of all kinds of heterogeneous data of business organizations in different industries, we propose a Heterogeneous Federated Graph Neural Network (HFGNN) approach. In this method, the distribution of heterogeneous business data of trans-border organizations is taken as subgraphs, and the sharing and circulation process among subgraphs is constructed as a statistically heterogeneous global graph through a central server. Each subgraph learns the corresponding personalized service model through local training to select and update the relevant subset of subgraphs with aggregated parameters, and effectively separates and combines topological and feature information among subgraphs. Finally, our simulation experimental results show that the proposed method has higher accuracy performance and faster convergence speed than existing methods.

Figures (7)

• Figure 1: Financial data trans-border sharing scenarios.
• Figure 2: Spatial graph structure: the nodes in the left graph denote the edge servers in different countries in the global trans-border flow of data, which denote the eight countries, such as China, Canada, France, Japan, Russia, United States, Brazil, Australia, etc.; the right graph denotes the graph structure formed by these eight countries in the dynamic process of trans-border data The right figure represents the graph structure of these eight countries in the dynamic process of trans-border data flow, where the nodes represent the edge servers in the countries, the edges represent the transmission paths between the edge servers in the trans-border data flow, and the thickness of the edges represents the size of the transmitted data volume in the data flow or sharing.
• Figure 3: Overview of the proposed model framework. (a, b) describes the original spatial graph among the eight countries or regions constructed based on their data provider and data demander; (c) illustrates the process of global graph dynamics evolution, which takes the data provider or demander as input and outputs a refined graph.
• Figure 4: Node state transfer process.
• Figure 5: Dynamic evolutionary dynamics of global maps.