Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· pp. 5568-5578· 0 citations· 13 references
Abstract
Subgraph federated learning (subgraph FL) enables collaborative graph neural network training without sharing raw graph data, but suffers from severe Non-IID distributions and structural fragmentation. In such settings, Non-IID distributions induce pronounced client specialization, where each client excels in a subset of nodes but remains insufficiently trained on others. However, existing similarity-based aggregation and distillation methods fail to balance client specialization and global generalization, often reinforcing dominant local representations and impairing generalization on under-represented nodes, or over-incorporating global knowledge and thereby disrupting well-optimized local specialization. In this paper, we propose FedDUA, a novel disagreement-aware and uncertainty-guided framework for subgraph FL. Specifically, FedDUA first models cross-client semantic disagreement via a lightweight semantic anchor graph and derives adaptive aggregation weights for reliable global federated knowledge. On the client side, FedDUA introduces an uncertainty-aware local-global semantic alignment mechanism that selectively reinforces representations of confident nodes while guiding uncertain nodes with aggregated global knowledge, thereby balancing local specialization and global generalization. Extensive experiments on six real-world datasets demonstrate that FedDUA consistently outperforms the state-of-the-art subgraph FL methods across varying numbers of clients. Further analyses validate the robustness and effectiveness of the proposed disagreement modeling and uncertainty-aware local-global semantic alignment strategies.
Heterogeneous graphs are a widely adopted data representation paradigm owing to their ability to model multiple node types and complex relationships uniformly. However, due to privacy and ownership restrictions, semantic-rich heterogeneous graphs are often stored by multiple participants, each of whom holds a subgraph. Federated heterogeneous subgraph learning seeks to aggregate knowledge from structurally diverse, decentralized subgraphs. This collaborative approach trains a more robust graph model without exposing raw data, thus overcoming the challenges of modeling complex relationships. In practical applications, however, the incompleteness and heterogeneity of subgraphs often result in degraded performance of the federated heterogeneous subgraph learning model. The method based on structural complementarity for addressing structural incompleteness ignores the semantic rules and node roles that are unique to heterogeneous graphs. For heterogeneous subgraphs, the information compression method based on category labels loses key high-order topological semantics. Therefore, we propose a new federated learning framework, federated heterogeneous subgraph learning with meta-path-guided role mapping and condensation (FedPRC). Specifically, based on the principle of structural equivalence, we propose a meta-path-guided virtual node generation method. This method enables each client to effectively infer and complete missing semantic links locally. Based on the evaluation of the importance of subgraph structure, we design a meta-path-guided information-aware subgraph condensation method. This method uses the condensation subgraph as an effective carrier for cross-client knowledge alignment and communication, while retaining the key topology and high-order paths. Our experiments on multiple real-world graph datasets demonstrate that this framework significantly outperforms existing federated graph learning baseline models in node classification tasks, validating its effectiveness.
Yanjin Cheng, Wenmin Li, Su-Lu Qin et al.· Journal of King Saud Univers...· 0 citations
This work proposes a FedIGL framework based on invariant learning, which effectively disrupts spurious correlations and further mines the invariant factors across different distributions, and proposes a novel Bi-Gradient Regularization strategy that introduces gradient constraints to guide the model in identifying client-agnostic and client-specific subgraph patterns for better graph representations.
Lingren Wang, Wenxuan Tu, Jiaxin Wang et al.· Neural Information Processin...· 2 citations
This work proposes a counterfactual multi-view framework to explicitly learn a structure masker for extracting consensus subgraphs and proposes a history-aware adaptive multi-channel aggregation mechanism to enable dynamic and personalized sharing driven by channel similarity while preventing collaborative oscillation.
Wan Zhang, Xiaoqian Jiang, Ye Wang et al.· 0 citations
FedTopo is proposed, a relation-level framework that encodes global knowledge as class relation topology, capturing how classes relate within each client rather than where they lie in feature space.
