Using AI to Enable Rapid Simulation of Extreme-Scale Physics Models (Facebook)

Speaker:
Kevin Carlberg, PhD, AI Research Science Manager at Facebook

Session Description:
Physics-based modeling and simulation has become indispensable across many applications in science, engineering, and entertainment, ranging from aircraft design to interacting with virtual objects. However, achieving truly high predictive fidelity in computational models necessitates fine spatiotemporal resolution, which can lead to extreme-scale models whose simulations consume months on thousands of computing cores. This constitutes a formidable computational barrier: the cost of truly high-fidelity simulations renders them impractical for important time-critical applications such as rapid design, control, or real-time interactive simulation. In this talk, I will present several advances that leverage cutting-edge AI techniques ranging from convolutional autoencoders to recurrent neural networks networks to overcome this barrier. In particular, these AI-infused methods produce low-dimensional counterparts to high-fidelity models called reduced-order models (ROMs) that exhibit 1) accuracy, 2) low cost, 3) physical-property preservation, 4) guaranteed generalization performance, and 5) error quantification, thus opening the door for extreme-scale physics models to be used in a new frontier of time-critical applications.

Follow and like us on social media: