We determine the exact minimax rate of a Gaussian sequence model under bounded convex constraints, purely in terms of the local geometry of the given constraint set $K$. Our main result shows that the minimax risk (up to constant factors) under the squared $\ell_2$ loss is given by $\epsilon^{*2} \wedge \operatorname{diam}(K)^2$ with \begin{align*} \epsilon^* = \sup \bigg\{\epsilon : \frac{\epsilon^2}{\sigma^2} \leq \log M^{\operatorname{loc}}(\epsilon)\bigg\}, \end{align*} where $\log M^{\operatorname{loc}}(\epsilon)$ denotes the local entropy of the set $K$, and $\sigma^2$ is the variance of the noise. We utilize our abstract result to re-derive known minimax rates for some special sets $K$ such as hyperrectangles, ellipses, and more generally quadratically convex orthosymmetric sets. Finally, we extend our results to the unbounded case with known $\sigma^2$ to show that the minimax rate in that case is $\epsilon^{*2}$.
翻译:我们确定一个高斯序列模型的精确微缩比例, 纯粹以设定约束值的本地几何来决定 $K$。 我们的主要结果显示, 平方 $\ ell_ 2$ 损失的最小最大风险( 直至恒定系数) 由 $\ epsilon=2 美元 =\ begin{ align} (K) =2$ =\ bigg ⁇ eepsilon :\ frac =2 = 2\ bigg = eepslon :\ frac = 2\ tgma2\\\\\ leq\ log M ⁇ operatorname{ { loc\\\\\\\\\\\\ epsilon\\\\ bigg}\ bigg_ 表示, = legal- squal- lax lax mexal laxal =s sex ex ex ex exmexmexmexmexmexmexmal 。 我们利用我们抽象结果, exal- ex- exal- ex- as mission lax lax lax lax lax lax lax lax lax lax laxk- slg- labs mex lap lax lax lax lax sl = lads mex lads sl = = ex sl ex sals ex = = = exg- sl exxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx