Polynomial-time near-optimal estimation over certain type-2 convex bodies

M. Neykov

Northwestern University, USA

We will discuss near (minimax) optimal computationally efficient estimators for constrained estimation over certain special convex constraints. The methodology is applicable to the problem of the constrained Gaussian sequence model, as well as constrained robust mean estimation and regression. Examples of sets over which this method can be applied include all well-balanced linearly transformed quadratically convex orthosymmetric sets.

Keywords: Constrained estimation, robust estimation.