FedHERO: A Federated Learning Approach for Node Classification Task on Heterophilic Graphs

TL;DR

This work tackles node classification in federated settings with heterophilic graphs, where differing neighbor distributions across clients degrade standard FL aggregation. It proposes FedHERO, a dual-channel GNN that jointly learns a global latent structure and client-specific local representations, sharing a structure-learner across clients to produce a common latent graph while retaining local graph information. Empirical results across semi-synthetic and real-world datasets show FedHERO consistently outperforms baselines, with advantages amplified under higher heterophily, faster convergence, and improved privacy. The approach offers a practical means to harness cross-client structural insights without sacrificing personalization or data confidentiality.

Abstract

Federated Graph Learning (FGL) empowers clients to collaboratively train Graph neural networks (GNNs) in a distributed manner while preserving data privacy. However, FGL methods usually require that the graph data owned by all clients is homophilic to ensure similar neighbor distribution patterns of nodes. Such an assumption ensures that the learned knowledge is consistent across the local models from all clients. Therefore, these local models can be properly aggregated as a global model without undermining the overall performance. Nevertheless, when the neighbor distribution patterns of nodes vary across different clients (e.g., when clients hold graphs with different levels of heterophily), their local models may gain different and even conflict knowledge from their node-level predictive tasks. Consequently, aggregating these local models usually leads to catastrophic performance deterioration on the global model. To address this challenge, we propose FedHERO, an FGL framework designed to harness and share insights from heterophilic graphs effectively. At the heart of FedHERO is a dual-channel GNN equipped with a structure learner, engineered to discern the structural knowledge encoded in the local graphs. With this specialized component, FedHERO enables the local model for each client to identify and learn patterns that are universally applicable across graphs with different patterns of node neighbor distributions. FedHERO not only enhances the performance of individual client models by leveraging both local and shared structural insights but also sets a new precedent in this field to effectively handle graph data with various node neighbor distribution patterns. We conduct extensive experiments to validate the superior performance of FedHERO against existing alternatives.

Figures (15)

• Figure 1: An example of financial transaction networks in two banks. The edges in the figure represent transaction records. Due to the diverse customer financial habits in distinct networks, customers (nodes) within the same risk classification (label) across different banks have diverse transactional relationship patterns (neighbor distribution).
• Figure 2: The framework of FedHERO. Blue boxes represent models in the $f_{local}$ channel, which are trained locally and personalized for each client. Yellow boxes denote models in the $f_{global}$ channel, shared among clients to train a structure learner for mutual benefit. On the right side of the figure, the model aggregation scheme in FedHERO is depicted.
• Figure 3: Performance of FedHERO and baselines on three real-world datasets.
• Figure 4: Convergence analysis of FedHERO and baselines on (left) Squirrel and (right) Chameleon datasets.
• Figure 5: (Left) Link inference attack (LIA) accuracies on FedHERO and FGL baseline methods. (Right) Performance of FedHERO and FGL baseline methods on noisy graph datasets.