Machine Learning for Simulation

The “Graph Guys”, as we like to call ourselves, focuses on the intersection of deep learning and physical sciences. We specialize in Geometric Deep Learning and Graph Neural Networks (GNNs), leveraging these architectures to model the complex, non-Euclidean relationships inherent in scientific data. This data includes irregular meshes for structural mechanics and dynamic particle interactions in fluid systems.

We believe that for ML to be truly transformative in engineering, it must understand the underlying principles of physics, adapt to changing environmental conditions, and provide reliable results in data-scarce regimes. By integrating symmetries, conservation laws, and multi-scale reasoning into our models, we are building the next generation of ML models that enable high-speed, high-fidelity simulations for applications like prototyping and design optimization.

A cornerstone of our team’s research is our close partnership with the Institute of Vehicle System Technology (FAST).

Key Areas

• Graph Network Simulators
• Learned Adaptive Meshing
• Physics-Consistent Dynamics