Scientific Machine Learning

Scientific Machine Learning (SciML) is an emerging field that lies at the intersection of physical modeling and simulation (based mostly on numerical partial differential equations) and data-driven machine learning (based mostly on neural networks and deep learning). By integrating these two different approaches (physics-based and data-based), SciML produces novel computational methods that inherit the attractive features of both approaches. In particular, SciML algorithms should have the following desirable properties: convergence estimates, robustness or generalizability, and computational efficiency.

The goal of this cluster is to develop robust, efficient and precise AI algorithms by merging physical modeling based on partial differential equations, physical laws, and data-driven ML methods. The cluster’s proposed novel methodologies are tested on problems relevant to society.

Cluster Members

• Lead PI: Beatrice Riviere (Computational Applied Mathematics and Operations Research, Rice University)
• Vladimir Braverman (Computer Science, Rice University)
• Matthias Heinkenschloss (Computational Applied Mathematics and Operations Research, Rice University)
• Lu Zhang (Computational Applied Mathematics and Operations Research, Rice University)

Collaborators

• Santiago Segarra (Electrical & Computer Engineering, Rice University)

Research Areas

• Graph neural networks
• Model-based deep learning
• Numerics-informed neural networks
• Reduced order models
• Parareal time integrators based AI
• Inverse problems

Upcoming Events

• SciML Workshop at the Energy HPC Conference: February 27, 2025

Updates will be posted to the following web page, managed by the cluster: https://sciml.rice.edu/home