Published on: August 2024
Enabling parametric sweeps on exascale AI-integrated simulations through federated learning
The project is in collaboration with Prof. Somdatta Goswami at Johns Hopkins University.
This NAIRR-Pilot project develops an innovative AI-integrated framework to enhance the efficiency and accessibility of exascale multiphysics simulations. The approach combines advanced AI techniques with traditional numerical methods to enable comprehensive parameter space exploration at unprecedented scales. Key features include:
- A novel spatial coupling mechanism between AI and numerical solvers
- Efficient communication techniques for distributed computing
- Adaptive learning methods for exascale-level simulations
The project aims to democratize access to exascale-level insights without compromising accuracy. It integrates pre-trained AI models with selective use of numerical solvers in critical sections, significantly reducing computational costs while maintaining high accuracy.
The framework is demonstrated through benchmarking against carbon capture processes and additive manufacturing problems. By enabling efficient parametric sweeps of exascale simulations, the project contributes to national priorities in high-performance computing and AI research, while also supporting educational initiatives to cultivate the next generation of STEM talent.