Sheaf-MPNN: A Library for Sheaf Neural Networks as Message Passing
What it is
This is a PyTorch Geometric library that implements Sheaf Neural Networks (SNNs) by reformulating sheaf diffusion as local message passing. It eliminates the need to construct global sparse block matrices, thereby reducing memory overhead and improving scalability for heterophilic graphs.
The problem it solves
Standard GNNs assume shared feature spaces leading to oversmoothing, while existing Sheaf Neural Network implementations require assembling large global sparse matrices, causing high memory and computational costs.
What you could use it for
- Training graph neural networks on heterophilic graphs where standard GNNs suffer from oversmoothing
- Comparing different sheaf neural network architectures (diagonal, orthogonal, low-rank maps) on node and graph-level tasks
- Implementing sheaf-based diffusion on heterogeneous and directed graphs within PyTorch Geometric
Why it matters
Provides a scalable and unified framework for sheaf neural networks, enabling efficient training on complex graph structures without the computational bottleneck of global Laplacian assembly.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.