Graph Neural Processes: Towards Bayesian Graph Neural Networks
This work addresses uncertainty estimation in graph-based deep learning for applications like edge imputation, but it appears incremental as it adapts existing neural process frameworks to graph data.
The paper tackles the problem of uncertainty quantification and handling dynamic-sized graphs in graph neural networks by introducing Graph Neural Processes (GNP), which output distributions over target points based on sparsely observed context points, though no concrete numerical results are provided.
We introduce Graph Neural Processes (GNP), inspired by the recent work in conditional and latent neural processes. A Graph Neural Process is defined as a Conditional Neural Process that operates on arbitrary graph data. It takes features of sparsely observed context points as input, and outputs a distribution over target points. We demonstrate graph neural processes in edge imputation and discuss benefits and drawbacks of the method for other application areas. One major benefit of GNPs is the ability to quantify uncertainty in deep learning on graph structures. An additional benefit of this method is the ability to extend graph neural networks to inputs of dynamic sized graphs.