Advanced Data Assimilation
High-fidelity simulation in the era of data. We infer the hidden state of turbulent and nonlinear physical systems from sparse, noisy measurements — combining adjoint methods, reduced-order models, and generative machine learning.
From partial observations to full physics
A unifying principle across mathematics, physics, and machine learning: complex, irreversible behavior in a low-dimensional observation space can be explained as the projection of simpler hyperbolic dynamics from a larger space. Our work turns that principle into practical inverse methods for fluids, sensing, and beyond.
Adjoint-based assimilation
Variational and preconditioned-adjoint methods that reconstruct turbulent flow fields and forcing from wall data, PIV, and sparse sensors.
Machine-learning ROMs
CNN auto-encoders, Fourier-invertible encoders, and Kolmogorov–Arnold networks for nonlinear reduced-order models of homogeneous and wall-bounded flows.
Generative inverse physics
Diffusion and Gaussian-process models that solve ill-posed inverse problems — pressure, sources, and dangerous-event detection — with quantified uncertainty.
Around the lab
Research areas →
Thirteen active directions in data assimilation, sensing, and inverse problems, supported by NSF and AFOSR.
People →
Ph.D., master's, and undergraduate researchers, plus collaborators and alumni.
Publications →
Recent work in JFM, Physics of Fluids, Physical Review Fluids, and AIAA / APS forums.
Education & Outreach →
FAVE — Fluids, AI & Visualization for Education: K–12 workshops, playable science, and AI-generated animations.
Other activities →
Workshops, the C.A.T. seminar series, and award-winning piano performance.
Contact & openings →
We welcome prospective undergraduate, master's, and Ph.D. students and visiting scholars.