Accurately modeling spatiotemporal fields and tuning parameters of physics-based models remain challenging tasks, particularly when dealing with sparse, irregular, and noisy sensor data. Existing modeling approaches often fail to adequately incorporate physical constraints, leading to predictions that are physically implausible or inconsistent with the underlying governing equations. This project aims to addresses these challenges through an integrated, physics-informed AI framework combining transformer-based field inference methods with adaptive parameter estimation from sparse sensor data.
This is an ongoing project, actively developing new extensions to handle multi-field systems, partial physics discovery, uncertainty quantification, and real-world deployment scenarios.
Ananth Balashankar (NYU / Google)
Lakshminarayanan Subramanian (Advisor, NYU)