Huan Zhang

I am interested in atmospheric blocking events, a quasi-stationary geopotential height anomaly in mid-latitude regions. For its quasi-stationarity, it is closely related to extreme weather, e.g. heat waves. Explainable AI followed by careful dynamics-based analysis proved to be powerful in revealing the mechanisms and advancing the prediction. 

Ground state and excited state estimation has been a forever focus for determining the properties of quantum systems. For the hermitianity of Hamitonian operator, ground staten and excited state can be solved as a variational problem. Variational Monte Carlo (VMC) solved the variational problem in parameter space. We aim to improve the generalizing error in small-probability regimes when neural network ansatzs are applied.

Time-dependent evolution equations need to be solved efficiently and accurately. I am also interested in numerical analysis of widely-used algorithms in machine learning and scientific computing.

Reduced-order modeling means two approaches to me: one is classical multi-scale modeling, building interpretable models directly from first- principles and physical mechanisms by making careful approximations, while the other is data-driven, seeking the low-dimensional manifold/dynamics to express the data structure. I would like to boost the power of the former by organically combining it with the latter.