ISCL members perform research at the intersection of data science, applied mathematics, and high-performance computing to enhance the understanding of complex multifidelity and multiphysics phenomena in various applications. In other words - we create science-based AI algorithms for applications in mechanical and aerospace engineering, Earth systems modeling, nuclear fusion, and more. ISCL has access to multiple HPC resources such as Bebop/Swing/Polaris/Aurora/Sophia (at Argonne), Gilbreth/Anvil (Purdue), and Perlmutter (NERSC).
An overview of our various research directions may be found in the following talks (starting oldest first): [1], [2], [3], [4], [5], [6], [7] . Further information about publications can be found on Google Scholar and our software contributions are available on Github. A collection of posters of our work is available here. ISCL eagerly welcomes possibilities for education, collaboration, and consulting! Feel free to reach out to us for any questions. We also run a scientific machine learning seminar series with leading researchers in our area of study - check it out here!
News
- Melissa Adrian wins an outstanding poster presentation award at the Purdue University Indianapolis Research Symposium - congrats!
- We welcome Dr. Sen Wang from the University of Notre Dame as a new postdoctoral scholar at ISCL. Sen previously worked on experimental and computational marine fog microphysics and will lead research into diffusion modeling for urban digital twins. Welcome!
- A manuscript led by former postdoc (and now faculty at UNC Charlotte) Dr. Xuyang Li is accepted to Neural Networks! This work studies how linear stability analysis may be used to determine safe behavior of a pretrained agent during deep reinforcement learning. Read more in the preprint here.
- Pleased to announce a new project supported by the Office of Naval Research for the multifidelity data assimilation of high-speed/hypersonic wall bounded flows. We will be using high fidelity simulation and experimental data to improve subgrid and wall models for large eddy simulations.
- Introducing a new preprint led by Dibyajyoti Chakraborty and Hojin Kim on adaptive placement of sensor locations for diffusion posterior sampling. We devise strategies to optimal place sensors such that diffusion-based surrogates of multiscale systems are provided improved information for data assimilation. Preprint is available here.
- Our paper on using an improved spectral loss function for improved forecasting of turbulent flows is accepted in the Journal of Computational Physics! Congratulations to Dibyajyoti Chakraborty who led this work in collaboration with Arvind Mohan at Los Alamos National Laboratory. A preprint can be found here.
- Please to announce a couple of new projects funded by DARPA (for agentic control of diffusion models) and DOE NEUP (for Graph Neural Network surrogates for data assimilation in nuclear reactors, in collaboration with TAMU and Idaho National Laboratory).
- Announcing a new preprint from our group, led by Haiwen Guan and collaborative with Prof. Ashesh Chattopadhyay at UC Santa Cruz, on deterministic and probabilistic downscaling of coarse-grained AI Climate emulators. In this work, we show that diffusion-based (probabilistic) downscaling models lead to improved recovery of fine-scaled spatial structures and temporal trends.
- See other archived news here.
Sponsors
