Home / Events / Statistical guarantees for neural operator surrogates
Statistics Seminar

Statistical guarantees for neural operator surrogates

With Sven Wang (EPFL)

Statistical guarantees for neural operator surrogates

In recent years, “operator learning” methodologies for constructing data-driven surrogates for non-linear operators are gaining widespread attention. We present statistical convergence results for the learning of such non-linear mappings in infinite-dimensional spaces, e.g. arising from PDEs, given noisy input-output pairs. We provide convergence results for least-squares-type empirical risk minimizers over general classes, in terms of their approximation properties and metric entropy bounds. This generalizes classical results from finite-dimensional nonparametric regression to an infinite-dimensional setting.

Assuming $G_0$ to be holomorphic, we prove algebraic (in the sample size $n$) convergence rates in this setting, thereby overcoming the curse of dimensionality. To illustrate the wide applicability, as a prototypical example we discuss the learning of the non-linear solution operator to a parametric elliptic partial differential equation, with an encoder-decoder based neural operator architecture.

Add to your calendar or Include in your list