Hector: An Efficient Programming and Compilation Framework for Implementing Relational Graph Neural Networks in GPU Architectures
Hector is a novel two-level intermediate representation and code generation framework that systematically addresses the performance and programming challenges of implementing relational graph neural networks (RGNNs) on GPU architectures.
Incorporating First-Order Partial Differential Equations into Graph Neural Networks for Enhanced Performance and Interpretability
This paper presents new Graph Neural Network models that incorporate two first-order Partial Differential Equations (PDEs) - the advection equation and the Burgers equation. These models effectively mitigate the over-smoothing problem in GNNs while maintaining comparable performance to higher-order PDE models.
Graph Neural Networks with Metastable Aggregation-Diffusion Dynamics
Aggregation-diffusion equations on graphs can exhibit metastable behavior, leading to clustered node representations that mitigate over-smoothing in graph neural networks.
DepWiGNN: A Novel Graph Neural Network for Multi-hop Spatial Reasoning in Text
DepWiGNN introduces a novel approach for multi-hop spatial reasoning in text, focusing on depth-wise propagation in graphs to capture long dependencies effectively.
DepWiGNN: A Novel Graph Neural Network for Multi-hop Spatial Reasoning in Text
Proposing a novel Depth-Wise Graph Neural Network (DepWiGNN) to address multi-hop spatial reasoning challenges by operating over the depth dimension of the graph.
