Uncertainty Modeling in Graph Neural Networks via Stochastic Differential Equations

Jun 1, 2023·
Richard Bergna
,
Felix Opolka
,
Pietro Liò
,
Jose Miguel Hernandez-Lobato
· 1 min read
Abstract
This research explores the application of stochastic differential equations to uncertainty modeling in graph neural networks (GNNs). The proposed approach enhances the expressiveness of GNNs by incorporating continuous stochastic processes in the representation learning process, which improves the network’s ability to handle uncertainty in graph-structured data.
Type
Publication
Preprint - Currently under review for ICLR 2024

This work provides a framework for leveraging stochastic differential equations to address uncertainty in graph neural networks, positioning it as a key approach for robust, uncertainty-aware models in structured data applications.

Uncertainty Quantification Graph Neural Networks Machine Learning