aureliayang / xecoslim

Acceleration of EcoSLIM using deep learning

XEcoSLIM

We build the deep learning model to further accelerate the GPU version EcoSLIM.

As the first step, we directly adopted
(1) A DNN model from fluid dynamics
(2) Neural ODE
Currently, we are training the DL model based on hillslope physical model and modifying the DL model to adapt water age simulations.

Demonstration

One year prediction of multi-point trajectories

Therefore, our work is easy to be applied to predict transport of plumes.

References

[1] Han M., Sane S., Johnson C. (2022). Exploratory lagrangian-based particle tracing using deep learning. Journal of Flow Visualization & Image Processing, 29(3).
[2] Chen R. , Rubanova Y., Bettencourt J., Duvenaud K. (2018). Neural ordinary differential equations. In Advances in Neural Information Processing Systems, 31.
[3] Sahoo S., Lu Y., Berger M. (2022). Neural Flow Map Reconstruction. Computer Graphics Forum, 41.
[4] Lu Y., Jiang K., Levine J. A., Berger M. (2021). Compressive Neural Representations of Volumetric Scalar Fields. Computer Graphics Forum, 40.