Title: Physics Informed Neural Networks for Accelerated Complex Flow Modeling and Discovery
Abstract: ROBIN-NET, a DOE Genesis Mission collaboration among The Ohio State University, Los Alamos National Laboratory (LANL), and Oak Ridge National Laboratory, aims to accelerate modeling and inverse inference of complex fluid flows through physics-informed AI. The project combines compact 3D Gaussian representations, diffusion models, and uncertainty quantification, with an emphasis on preserving physics during creation of the latent training data and throughout prediction. This talk will introduce the project and describe how LANL is establishing its benchmarking and validation foundation. We are assembling DOE-relevant fluid-flow datasets, assessing which conservation laws and physical constraints can be evaluated and how well the data satisfy them, and identifying baseline models for systematic comparison. These efforts will help distinguish data limitations from model errors and establish how advances in physics consistency, predictive accuracy, and computational efficiency will be measured.
Bio: Ayan Biswas is a senior scientist at Los Alamos National Laboratory, where his research spans scientific machine learning, foundation models, scientific visualization, and multimodal and agentic AI. He co-leads the foundation-model research thrust within LANL’s ArtIMis initiative, serves as LANL principal investigator for the DOE Genesis Mission project ROBIN-NET, and is co-PI of the new LDRD Exploratory Research project “Physics-Informed AI Surrogates for Fusion Fuel Cycle Component Design.” His leadership experience includes serving as co-PI of the LDRD Directed Research project “In-Situ Inference,” leading agentic visualization research within Advanced Simulation and Computing (ASC), and serving as deputy project lead for LLM-based code translation and federated learning efforts. He also contributed to the Exascale Computing Project’s ALPINE effort through research in in situ visualization and scalable data reduction. He earned his Ph.D. from The Ohio State University, specializing in data analysis and visualization.