ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Two approaches to generalize Stein's method to non-Gaussian distributions are the following: \begin{itemize} \item[(i)] Stein coefficients: for a random variable YY, the \emph{Stein coefficient} T(Y)T(Y) satisfies

E(Yf(Y))=E(Tf′(Y)).\mathbb{E}(Y f(Y)) = \mathbb{E}(T f'(Y)).

\item[(ii)] Zero-biasing: for YY of mean 0 and variance σ2\sigma^2, compute the Y∗Y^* that satisfies

E(Yf(Y))=σ2Ef′(Y∗).\mathbb{E}(Yf(Y)) = \sigma^2 \mathbb{E} f'(Y^*).

Let Y1,…,Yd∼i.i.d.YY_1, \ldots, Y_d \overset{i.i.d.}\sim Y, and let Y‾=(Y1,…,Yd)\overline{Y} = (Y_1, \ldots, Y_d). Suppose that EYi=0\mathbb{E} Y_i = 0, and let X=Y+θX = Y + \theta. Here we will assume YY is a known distribution while θ\theta is unknown.

Consider estimators θ^\hat{\theta} of Θ\Theta of the form θ^=X+h(X)\hat{\theta} = X + h(X). Recall that SURE for Gaussian random variables gives the unbiased estimate of risk

r^h=d+2∇⋅h(X)+∥h(X)∥2.\hat{r}_h = d + 2\nabla \cdot h(X) + \|h(X)\|^2.

Calculate the bias

ed:=E(r^h−E∥θ^−θ∥2)e_d:= \mathbb{E}(\hat{r}_h - \mathbb{E} \|\hat{\theta} - \theta\|^2)

using Stein coefficients or zero-biasing. Does edd\frac{e_d}{d} converge to 0 as d→∞d \rightarrow \infty? Can we apply this to wavelet shrinkage? \end{itemize}

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.