Credal Graph Neural Networks

2025-12-02

AI Summary

Existing GNN-based uncertainty quantification methods—such as Bayesian inference and ensemble learning—often fail to reliably characterize epistemic uncertainty in out-of-distribution (OOD) node classification, especially on heterogeneous graphs. To address this, we propose Credible Graph Neural Networks (CGNN), the first framework to introduce credal set modeling into graph learning, enabling explicit, set-valued predictions that capture epistemic uncertainty. CGNN innovatively leverages inter-layer information complementarity in graph message passing to design a heterogeneity-aware credible learning strategy, and theoretically and empirically reveals the critical role of graph homophily in uncertainty estimation. Evaluated on multiple heterogeneous graph benchmarks, CGNN significantly improves calibration and robustness of uncertainty estimates under OOD conditions, achieving state-of-the-art performance in both uncertainty quantification and downstream classification tasks.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Graphical Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks

Abstract

Uncertainty quantification is essential for deploying reliable Graph Neural Networks (GNNs), where existing approaches primarily rely on Bayesian inference or ensembles. In this paper, we introduce the first credal graph neural networks (CGNNs), which extend credal learning to the graph domain by training GNNs to output set-valued predictions in the form of credal sets. To account for the distinctive nature of message passing in GNNs, we develop a complementary approach to credal learning that leverages different aspects of layer-wise information propagation. We assess our approach on uncertainty quantification in node classification under out-of-distribution conditions. Our analysis highlights the critical role of the graph homophily assumption in shaping the effectiveness of uncertainty estimates. Extensive experiments demonstrate that CGNNs deliver more reliable representations of epistemic uncertainty and achieve state-of-the-art performance under distributional shift on heterophilic graphs.

Problem

• Develop credal graph neural networks for uncertainty quantification
• Address out-of-distribution node classification using set-valued predictions
• Enhance epistemic uncertainty reliability on heterophilic graphs

Innovation

• Extends credal learning to graph neural networks for set-valued predictions
• Leverages layer-wise message passing for uncertainty quantification in graphs
• Enhances epistemic uncertainty representation under distributional shift