Scientific Machine Learning for Spatio-temporal Predictions

Topic 1:Differentiable Neural Solver and GPU Computing

Aiming to develop next-gen computational modeling framework by combining classic numerical method, deep learning techniques, and data via differentiable programming and GPU computing — enabling efficient forward/inverse modeling and data assimilation.

Topic 2:Advance deep learning techniques

We are pushing the boundary of deep learning architecture, data-driven techniques for learning spatiotemporal system – graph neural networks, sequence2sequence networks, transformer, diffusion models et.

Topic 3:Physics-informed, PDE-regularized Machine/Deep Learning

Using residual of governing PDEs to regularized or drive machine/deep learning model training, e.g., physics-informed neural networks, B-PINN, physics-informed CNN, and FEM-GNN etc.