Problem
Research questions and friction points this paper is trying to address.
Out-of-Distribution Generalization
Graph Neural Networks
Self-Supervised Learning
Node Classification
Distribution Shift
Complementary Supervised and Self-Supervised Representations for Out-of-Distribution Graph Learning
🤖 AI Summary
This study addresses the underutilization of information in out-of-distribution (OOD) generalization for graph neural networks, caused by the disconnection between supervised and self-supervised representation objectives. To bridge this gap, we propose the Co-Train and Dual-Space Retrieval frameworks. The former adaptively integrates complementary signals through joint training, while the latter achieves two-stage synergy during inference via confidence-aware fusion, supporting independent parameterization of heterogeneous encoders. This work provides the first systematic validation of the complementarity between self-supervised learning and supervised OOD generalization. Experimental results demonstrate that the proposed method significantly outperforms strong supervised baselines across four benchmark datasets, achieving robust OOD node classification in diverse scenarios, including cross-domain temporal shifts.
📝 Abstract
Out-of-distribution (OOD) generalization remains challenging for graph neural networks (GNNs), as graph distributions can vary substantially across time and domains. Supervised and self-supervised graph representation learning are guided by distinct objectives and offer different perspectives on graph representations. In this work, we study whether self-supervised representations (SSL) can provide complementary signals to improve supervised OOD node classification. We develop two backbone-agnostic frameworks that exploit such information at different stages of learning and prediction. Co-Train jointly learns supervised and SSL representations and adaptively integrates them during training, while Dual-Space Retrieval performs non-parametric prediction in the two representation spaces and combines their predictions through confidence-aware fusion at inference time. The supervised and SSL encoders are separately parameterized and need not share the same GNN architecture.
We evaluate multiple GNN backbones and two distinct SSL objectives, DGI and GRACE, on four graph benchmarks spanning temporal and cross-domain distribution shifts. Extensive experiments show that Co-Train consistently outperforms strong supervised OOD baselines, while Dual-Space Retrieval achieves competitive performance as a flexible non-parametric alternative. Results across different backbones and SSL objectives, together with representation analyses and ablations, demonstrate that SSL representations provide complementary information to supervised representations and can improve OOD node classification across diverse settings.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Out-of-Distribution Graph Learning
Self-Supervised Representation
Graph Neural Networks
Node Classification
Complementary Fusion
🔎 Similar Papers
No similar papers found.
