Abstract: Scientific machine learning (SciML) is becoming a fundamental part of research, development, and discovery workflows for scientists across many domains such as computational fluid dynamics, materials characterization, climate and weather modeling, and computer-aided engineering. With the increase in use of AI and ML in these domains also comes an increase in the number of resources available for developing models, and a plethora of datasets to use as starting points for model training. While many efforts are segmented and tailor-made for specific use cases, projects such as PhysicsNeMo, The Well, and Proxima Fusion's ConStellaration dataset are great examples of community efforts to bridge the gap between siloed SciML research and collaborative community projects. In this tutorial, PhysicsNeMo is used to expand the scope of community models and datasets from The Well and the ConStellaration Challenge by leveraging pre-trained physics-informed machine learning models, community accessible datasets, and the robust SciML framework from PhysicsNeMo.
Bio: Carmelo Gonzales is a solutions architect at NVIDIA with a background in applied scientific AI and machine learning in HPC environments. Before joining NVIDIA, he worked on AI-enabled material discovery, AI-accelerated simulations, and built open software frameworks for the scientific research community. His current work is on enabling R&D teams to adopt and apply scientific AI/ML to problems in physics, with a focus on accelerated and scalable workflows.
