Abstract
Subgraph federated learning (subgraph-FL) is a new distributed paradigm thatfacilitates the collaborative training of graph neural networks (GNNs) bymulti-client subgraphs. Unfortunately, a significant challenge of subgraph-FLarises from subgraph heterogeneity, which stems from node and topologyvariation, causing the impaired performance of the global GNN. Despite variousstudies, they have not yet thoroughly investigated the impact mechanism ofsubgraph heterogeneity. To this end, we decouple node and topology variation,revealing that they correspond to differences in label distribution andstructure homophily. Remarkably, these variations lead to significantdifferences in the class-wise knowledge reliability of multiple local GNNs,misguiding the model aggregation with varying degrees. Building on thisinsight, we propose topology-aware data-free knowledge distillation technology(FedTAD), enhancing reliable knowledge transfer from the local model to theglobal model. Extensive experiments on six public datasets consistentlydemonstrate the superiority of FedTAD over state-of-the-art baselines.