B1088
Title: Challenges of experiment design in high-dimensional spaces
Authors: Mojmir Mutny - ETH Zurich (Switzerland) [presenting]
Abstract: Experiment design optimizes resource allocation to accurately estimate the quantity of interest within a prescribed measurement model. Delving into the challenges presented when applying this in non-parametric settings, namely within reproducing kernel Hilbert spaces. The challenges come in two types: computational and statistical. To overcome them, two tractable assumptions are introduced on the Hilbert: additive and projection pursuit assumptions. The assumptions cause estimation to be of low complexity but also optimization of allocations becomes likewise tractable. It is especially important for my-optic strategies that try to estimate the maximizer of an unknown function.
