Johannes Jakubik

A learned mapping can replace QAOA's expensive per-instance angle optimization, yielding near-identical results orders of magnitude faster and generalizing to new problem structures.

Architectural choices, not model identity, drive performance differences in Geospatial Foundation Models, challenging conventional ranking methods.

Meter-scale weather, previously thought irreducibly chaotic, actually contains a statistically recoverable component, unlocking a new frontier for high-resolution weather modeling.