Double the fun with two seminars this week!
Speaker: Björn Lütjens
Human Systems Laboratory
Speeding up climate models with Machine Learning
Abstract:
The world has less than 280GtCO2e emissions left to limit the global temperature increase to the 1.5°C Paris goal (IPCC, 2018). Policy and decision makers rely on climate models to explore decisions and calculate climate impacts and risks. Running a single climate model, however, can take 10days to run on a 5000GPU node supercomputer, using the same electricity a coal power plant generates in 1hour. Clearly this is too expensive to run climate models multiple times for exploring policy decisions or compute climate risks. The computational expense arises from numerical solvers that have to repeatedly solve partial differential equations (PDEs) for each choice of parameters, mesh, or forcing term. Novel methods from physics-informed deep learning promise to create accurate and fast copies, called surrogates, of numerical solvers that are 1−2 orders of magnitude faster. In this talk I will present results from building a x100-faster coastal flood model, a faster el Nino/la Nina prediction, and preliminary results for leveraging neural networks for a fast fundamental multi-scale PDE solver and uncertainty quantification.
Speaker: Ferrous Ward
Human Systems Laboratory
Virtual Reality to Support Mission Operations and Enhance Planetary Exploration Situational Awareness
Abstract:
Virtual Reality (VR) presents the opportunity to use digital environments to assist in reactive mission planning / data analysis, for mission control and to enhance situational awareness for astronauts on exploration missions. There are several core concepts that an immersive digital environment brings to a decision support system: telepresence, real-time data visualization, interactivity, and real-time data analysis. This then gives high potential to use the virtual environments to enhance remote planetary science operations and lay the foundation for a “standard interface” for VR planetary science.