pyt-team/TopoModelX

What it solves

TopoModelX (TMX) provides tools to implement topological neural networks (TNNs) for science and engineering. It allows researchers to go beyond standard graph-based deep learning by working with topological domains, such as simplicial complexes, to capture higher-order interactions in data.

How it works

TMX is a Python module that integrates with PyTorch. It uses topological structures like simplicial complexes and their associated matrices (such as up and down Laplacians and incidence matrices) to define neural network layers. For example, it includes implementations of Simplicial Attention Networks (SAN), which can process features on nodes and edges simultaneously to perform forward passes on topological data.

Who it’s for

It is designed for researchers and engineers in science and engineering who need to apply deep learning to data structured as topological complexes rather than simple graphs.

Highlights

• Simplicial Attention Networks (SAN): Implements specialized layers for topological deep learning.
• PyTorch Integration: Built as a PyTorch module for efficient training and training workflows.
• Part of TopoX: Integrates with the broader TopoX suite of packages for machine learning on topological domains.
• Sparse Matrix Support: Utilizes sparse matrices for efficient handling of topological Laplacians.