hi, i'm ali can — a postdoctoral researcher working on machine learning for fluid dynamics
and weather prediction at the helmholtz-zentrum hereon, within the model-driven machine learning group.
prior to arriving there, i completed my phd in aerospace engineering
at the university of arizona, where i worked on nonlocal physics-informed machine learning algorithms.
beyond my academic pursuits, i have a keen interest in linguistics and cinema.
research
ai-based weather and s2s prediction
lately, i've been working on data-driven probabilistic weather prediction models — mostly graph-based and trained on era5. my aim is to address some of the persistent issues with ai-based weather prediction, such as long-term stability and anomaly prediction. beating the state of the art along the way would be a nice bonus.
pde emulation with physical constraints
i recently published a paper on this topic at neurips examining the impact of resolution coarsening on pde emulation accuracy and the role of physical constraints in latent-space models. i'm now working more on differentiable physics models — there is a lot to explore, but i try to balance solving real problems with satisfying my curiosity. i guess i also like to see how ml models interact with physical solvers.
data assimilation and ml-based emulators
i'm fairly new to this field, but i enjoy exploring the intersection of da and ml-based emulators. having adjoints for free in variational da is appealing, and a couple of experiments have produced some unintuitive results — giving me a gut feeling that there's something interesting here worth exploring.