SBI4C is an interdisciplinary research project that develops new methodologies for parameter inference in large-scale physical simulators, with climate science as its central application domain. It is primarily funded by the MIAI cluster in Grenoble, France.
Goals
The overarching goal of SBI4C is to extend simulation-based inference methods to large-scale models such as climate simulators that capture complex ocean–atmosphere dynamics. Our main use-case is the important problem of tuning (i.e. calibration) of parametrizations of sub-grid processes in climate models.
More with less
Design SBI methods that deliver robust results despite few expensive simulator runs.
Better approximators
Tailored SBI architectures beyond off-the-shelf deep models.
Fast surrogates
Leveraging emulators for fast, reliable SBI.
Dissemination
Organization of workshops, summer school, and a hackathon.
