This is a living page of project ideas I am developing, each with a running log of relevant literature and my notes on it. If you are a student looking for a project, or a researcher interested in collaborating on any of these directions, please write to me — the notes below should give you a sense of where each idea stands.
Guiding Principle: What makes a good problem statement?
A good problem statement is essential for guiding the research process and includes some key elements. It should:
- Clearly define the core issue that the research aims to address.
- Provide context and background.
- Be specific and set the boundaries and scope of the work.
- Be something that can be realistically addressed.
- Justify the need for the research.
Uncertainty quantification for neural data assimilation
OpenEnd-to-end learned data assimilation frameworks such as 4DVarNet produce state reconstructions, but attaching trustworthy uncertainties to those reconstructions is an open problem. This project explores ensemble-based approaches — in particular spatiotemporally coherent perturbations — to turn a deterministic neural assimilation system into one that produces calibrated ensembles.
Literature & running notes
Spatiotemporal Coherent Displacements for Ensemble-Based Neural Data Assimilation
Our accepted EDS paper — the starting point for this project. The displacement-based perturbations give coherent ensembles; the open question is calibration: how do we verify and improve the spread–skill relationship in the learned setting?
Learning Variational Data Assimilation Models and Solvers (Fablet et al.)
The 4DVarNet reference. Read this first to understand the bi-level structure (inner assimilation solver, outer learned components) that any UQ scheme must respect.
More project ideas will appear here as the notes mature. If one of these resonates with you and you want to spend time learning and co-developing it further— reach out.
– Upcoming conference venues – https://metasysid.github.io/