2pm at NORCE (building F, floor 3).
Talk by Patrick N. Raanes (NORCE).


Least-square linear regression (LLS) is hiding (in plain sight?) in ensemble methods. By virtue of Stein’s identity, it can be shown to relate to average sensitivity,

\[\mathbf{Y} \tilde{\mathbf{X}}^{+} = \bar{\nabla}\! f \xrightarrow[N \rightarrow \infty]{} 𝔼 \nabla\! f(\mathbf{x}) \,,\]

without the awkward use of Taylor expansions.

The LLS perspective on ensemble methods can also be used to unify the many flavours of the analysis updates (Perturbed, EAKF, ETKF [subspace]), an effortless exercise thanks to its very own chain rule.