Structural differences harm federated graph neural network accuracy
Structural Negative Transfer in Federated Graph Neural Networks: Diagnosis, Causal Investigation, and the Limits of Divergence-Aware Mitigation
Machine LearningDistributed, Parallel, and Cluster Computing
Summary
Federated learning lets many users train a shared model without sharing their data, usually by combining their local updates. This works well when their data are similar, but the authors found a new problem when the data are graphs with different structures. In these cases, a client with a unique graph shape can suffer a large drop in model accuracy. The authors explored various ways to measure and fix this problem but found only a weak signal related to node connectivity and no clear solution.
What this means in practice
- •For machine learning engineers: Identify when structural differences in graph data may reduce model accuracy in federated graph neural networks.
- •For network analysts: Assess the impact of graph topology variation on collaborative learning performance across multiple networks.
Tested on one dataset.
Authors
Chethana Prasad Kabgere, Shylaja SS
Abstract
Federated learning lets multiple participants train a shared model without pooling raw data, by exchanging locally trained model updates instead. Federated averaging assumes that averaging local models is a reasonable way to solve one shared problem when participants' data are broadly similar. Work on non-IID federated learning has shown that this assumption can withstand differences in label and feature distributions. We ask whether it survives a different strain specific to graph neural networks, where client graphs differ not in label or feature distribution but in structure itself, requiring the same shared weights to operate over fundamentally different topologies. We call the resulting harm structural negative transfer. In a federation of real citation networks and synthetic structural proxies, a structurally atypical client lost more than half its achievable accuracy simply by joining. In an initial six-client federation, two label-free structural statistics computable before training were strongly associated with this harm. Expanding to twenty clients showed that degree divergence remained associated with harm, although more weakly, and survived removal of domain contrast. Spectral divergence did not replicate, which we trace to a confound caused by the composition of the reference pool used for leave-one-out statistics. A causal intervention isolating topology found no significant effect. A degree-normalization mechanism held across twenty-four seeds but did not explain the harm when corrected. The best of five candidate fixes beat a tuned baseline only until a matched, structurally blind control was applied, after which the gain disappeared. What survives is a modest, partially replicated, degree-specific signal that is not yet a validated predictor at scale.