Search papers, labs, and topics across Lattice.
This paper introduces Quantile-free Prediction Interval GNN (QpiGNN), a novel framework for uncertainty quantification in GNNs that directly optimizes for coverage and interval width without needing quantile inputs or post-processing. QpiGNN uses a dual-head architecture trained with a quantile-free joint loss function, enabling efficient training and robust prediction intervals. Empirical results across 19 datasets demonstrate that QpiGNN achieves 22% higher coverage and 50% narrower intervals compared to existing methods, while also exhibiting robustness to noise and structural shifts.
GNN uncertainty just got a whole lot easier: QpiGNN delivers better coverage and tighter intervals without the quantile gymnastics.
Uncertainty quantification (UQ) in graph neural networks (GNNs) is crucial in high-stakes domains but remains a significant challenge. In graph settings, message passing often relies on strong assumptions such as exchangeability, which are rarely satisfied in practice. Moreover, achieving reliable UQ typically requires costly resampling or post-hoc calibration. To address these issues, we introduce Quantile-free Prediction Interval GNN (QpiGNN), a framework that builds on quantile regression (QR) to enable GNN-based UQ by directly optimizing coverage and interval width without requiring quantile inputs or post-processing. QpiGNN employs a dual-head architecture that decouples prediction and uncertainty, and is trained with label-only supervision through a quantile-free joint loss. This design allows efficient training and yields robust prediction intervals, with theoretical guarantees of asymptotic coverage and near-optimal width under mild assumptions. Experiments on 19 synthetic and real-world benchmarks show QpiGNN achieves average 22\% higher coverage and 50\% narrower intervals than baselines, while ensuring efficiency and robustness to noise and structural shifts.