Abstract: The simulation of physics is foundational for all fields of science and industry but is typically bottle-necked by expensive computation. Many systems of scientific and industrial interest are computationally laborious to sufficiently resolve. With the recent advances in the machine learning field, modern deep neural network architectures have become increasingly more powerful and accessible than their predecessors developed over two decades ago. This has ushered in a new wave of physics and data-driven modeling for physical systems. NVIDIA Modulus is a neural network framework that blends the power of physics and partial differential equations (PDEs) with AI to build more robust models for better analysis.
In this presentation, we will discuss the implementation and applications of Modulus to model various physical systems ranging from heat sink optimization to global weather forecasting. Key features and design decisions used to generalize/scale physics-based machine learning will also be presented. With NVIDIA Modulus, we aim to provide researchers and industry specialists with various tools that will help accelerate the development of such deep learning models for various scientific disciplines.