### 统计代写|统计推断代写Statistical inference代考|MAST90100

statistics-lab™ 为您的留学生涯保驾护航 在代写统计推断Statistical inference方面已经树立了自己的口碑, 保证靠谱, 高质且原创的统计Statistics代写服务。我们的专家在代写统计推断Statistical inference代写方面经验极为丰富，各种代写统计推断Statistical inference相关的作业也就用不着说。

• Statistical Inference 统计推断
• Statistical Computing 统计计算
• (Generalized) Linear Models 广义线性模型
• Statistical Machine Learning 统计机器学习
• Longitudinal Data Analysis 纵向数据分析
• Foundations of Data Science 数据科学基础

## 统计代写|统计推断代写Statistical inference代考|Signal Recovery Problem

One of the basic problems in Signal Processing is the problem of recovering a signal $x \in \mathbf{R}^{n}$ from noisy observations
$$y=A x+\eta$$
of a linear image of the signal under a given sensing mapping $x \mapsto A x: \mathbf{R}^{n} \rightarrow \mathbf{R}^{m}$; in (1.1), $\eta$ is the observation error. Matrix $A$ in (1.1) is called sensing matrix.
Recovery problems of the outlined types arise in many applications, including, but by far not reducing to,

• communications, where $x$ is the signal sent by the transmitter, $y$ is the signal recorded by the receiver, and $A$ represents the communication channel (reflecting, e.g., dependencies of decays in the signals’ amplitude on the transmitter-receiver distances); $\eta$ here typically is modeled as the standard (zero mean, unit covariance matrix) $m$-dimensional Gaussian noise; ${ }^{1}$
• image reconstruction, where the signal $x$ is an image – a $2 \mathrm{D}$ array in the usual photography, or a 3D array in tomography-and $y$ is data acquired by the imaging device. Here $\eta$ in many cases (although not always) can again be modeled as the standard Gaussian noise;
• linear regression, arising in a wide range of applications. In linear regression, one is given $m$ pairs “input $a^{i} \in \mathbf{R}^{n \text { ” }}$ to a “black box,” with output $y_{i} \in \mathbf{R}$. Sometimes we have reason to believe that the output is a corrupted by noise version of the “existing in nature,” but unobservable, “ideal output” $y_{i}^{*}=x^{T} a^{i}$ which is just a linear function of the input (this is called “linear regression model,” with inputs $a^{i}$ called “regressors”). Our goal is to convert actual observations $\left(a^{i}, y_{i}\right), 1 \leq i \leq m$, into estimates of the unknown “true” vector of parameters $x$. Denoting by $A$ the matrix with the rows $\left[a^{i}\right]^{T}$ and assembling individual observations $y_{i}$ into a single observation $y=\left[y_{1} ; \ldots ; y_{m}\right] \in \mathbf{R}^{m}$, we arrive at the problem of recovering vector $x$ from noisy observations of $A x$. Here again the most popular model for $\eta$ is the standard Gaussian noise.

## 统计代写|统计推断代写Statistical inference代考|Parametric and nonparametric cases

Recovering signal $x$ from observation $y$ would be easy if there were no observation noise $(\eta=0)$ and the rank of matrix $A$ were equal to the dimension $n$ of the signals. In this case, which arises only when $m \geq n$ (“more observations than unknown parameters”), and is typical in this range of $m$ and $n$, the desired $x$ would be the unique solution to the system of linear equations, and to find $x$ would be a simple problem of Linear Algebra. Aside from this trivial “enough observations, no noise” case, people over the years have looked at the following two versions of the recovery problem:

Parametric case: $m \gg n, \eta$ is nontrivial noise with zero mean, say, standard Gaussian. This is the classical statistical setup with the emphasis on how to use numerous available observations in order to suppress in the recovery, to the extent possible, the influence of observation noise.

Nonparametric case: $m \ll n .^{2}$ If addressed literally, this case seems to be senseless: when the number of observations is less that the number of unknown parameters, even in the noiseless case we arrive at the necessity to solve an undetermined (fewer equations than unknowns) system of linear equations. Linear Algebra says that if solvable, the system has infinitely many solutions. Moreover, the solution set (an affine subspace of positive dimension) is unbounded, meaning that the solutions are in no sense close to each other. A typical way to make the case of $m \ll n$ meaningful is to add to the observations (1.1) some a priori information about the signal. In traditional Nonparametric Statistics, this additional information is summarized in a bounded convex set $X \subset \mathbf{R}^{n}$, given to us in advance, known to contain the true signal $x$. This set usually is such that every signal $x \in X$ can be approximated by a linear combination of $s=1,2, \ldots, n$ vectors from a properly selected basis known to us in advance (“dictionary” in the slang of signal processing) within accuracy $\delta(s)$, where $\delta(s)$ is a function, known in advance, approaching 0 as $s \rightarrow \infty$. In this situation, with appropriate $A$ (e.g., just the unit matrix, as in the denoising problem), we can select some $s \leqslant m$ and try to recover $x$ as if it were a vector from the linear span $E_{s}$ of the first $s$ vectors of the outlined basis $[54,86,124,112,208]$. In the “ideal case,” $x \in E_{s}$, recovering $x$ in fact reduces to the case where the dimension of the signal is $s \ll m$ rather than $n \gg m$, and we arrive at the well-studied situation of recovering a signal of low (compared to the number of observations) dimension. In the “realistic case” of $x \delta(s)$-close to $E_{s}$, deviation of $x$ from $E_{s}$ results in an additional component in the recovery error (“bias”); a typical result of traditional Nonparametric Statistics quantifies the resulting error and minimizes it in $s[86,124,178,222,223,230,239]$. Of course, this outline of the traditional approach to “nonparametric” (with $n \gg m$ ) recovery problems is extremely sketchy, but it captures the most important fact in our context: with the traditional approach to nonparametric signal recovery, one assumes that after representing the signals by vectors of their coefficients in properly selected base, the $n$-dimensional signal to be recovered can be well approximated by an $s$-sparse (at most $s$ nonzero entries) signal, with $s \ll n$, and this sparse approximation can be obtained by zeroing out all but the first $s$ entries in the signal vector. The assumption just formulated indeed takes place for signals obtained by discretization of smooth uni- and multivariate functions, and this class of signals for several decades was the main, if not the only, focus of Nonparametric Statistics.

## 统计代写|统计推断代写Statistical inference代考|Compressed Sensing via ℓ1 minimization: Motivation

In principle there is nothing surprising in the fact that under reasonable assumption on the $m \times n$ sensing matrix $A$ we may hope to recover from noisy observations of $A x$ an $s$-sparse signal $x$, with $s \ll m$. Indeed, assume for the sake of simplicity that there are no observation errors, and let $\operatorname{Col}{j}[A]$ be $j$-th column in $A$. If we knew the locations $j{1}<j_{2}<\ldots<j_{s}$ of the nonzero entries in $x$, identifying $x$ could be reduced to solving the system of linear equations $\sum_{\ell=1}^{s} x_{i_{\ell}} \operatorname{Col}_{j \ell}[A]=y$ with $m$ equations and $s \ll m$ unknowns; assuming every $s$ columns in $A$ to be linearly independent (a quite unrestrictive assumption on a matrix with $m \geq s$ rows), the solution to the above system is unique, and is exactly the signal we are looking for. Of course, the assumption that we know the locations of nonzeros in $x$ makes the recovery problem completely trivial. However, it suggests the following course of action: given noiseless observation $y=A x$ of an s-sparse signal $x$, let us solve the combinatorial optimization problem
$$\min {z}\left{|z|{0}: A z=y\right},$$
where $|z|_{0}$ is the number of nonzero entries in $z$. Clearly, the problem has a solution with the value of the objective at most $s$. Moreover, it is immediately seen that if every $2 s$ columns in $A$ are linearly independent (which again is a very unrestrictive assumption on the matrix $A$ provided that $m \geq 2 s$ ), then the true signal $x$ is the unique optimal solution to $(1.2)$.
What was said so far can be extended to the case of noisy observations and “nearly $s$-sparse” signals $x$. For example, assuming that the observation error is “uncertainbut-bounded,” specifically some known norm $|\cdot|$ of this error does not exceed a given $\epsilon>0$, and that the true signal is s-sparse, we could solve the combinatorial optimization problem
$$\min {z}\left{|z|{0}:|A z-y| \leq \epsilon\right} .$$
Assuming that every $m \times 2 \mathrm{~s}$ submatrix $\bar{A}$ of $A$ is not just with linearly independent columns (i.e., with trivial kernel), but is reasonably well conditioned,
$$|\bar{A} w| \geq C^{-1}|w|_{2}$$
for all ( $2 s)$-dimensional vectors $w$, with some constant $C$, it is immediately seen that the true signal $x$ underlying the observation and the optimal solution $\widehat{x}$ of (1.3) are close to each other within accuracy of order of $\epsilon:|x-\widehat{x}|_{2} \leq 2 C \epsilon$. It is easily seen that the resulting error bound is basically as good as it could be.

## 统计代写|统计推断代写Statistical inference代考|Signal Recovery Problem

• 通讯，在哪里X是发射机发送的信号，是是接收器记录的信号，并且一个表示通信信道（反映，例如，信号幅度衰减对发射机-接收机距离的依赖性）；这这里通常被建模为标准（零均值，单位协方差矩阵）米-维高斯噪声；1
• 图像重建，其中信号X是一个图像——一个2D通常摄影中的阵列，或断层扫描中的 3D 阵列 – 和是是成像设备获取的数据。这里这在许多情况下（尽管并非总是如此）可以再次建模为标准高斯噪声；
• 线性回归，在广泛的应用中出现。在线性回归中，给出一个米对“输入一个一世∈Rn ” 到一个“黑匣子”，输出是一世∈R. 有时我们有理由相信输出是“存在于自然界”但不可观察的“理想输出”的噪声版本是一世∗=X吨一个一世这只是输入的线性函数（这称为“线性回归模型”，输入一个一世称为“回归器”）。我们的目标是转换实际观察结果(一个一世,是一世),1≤一世≤米, 估计未知的“真实”参数向量X. 表示一个具有行的矩阵[一个一世]吨并收集个人观察结果是一世一次观察是=[是1;…;是米]∈R米，我们得到了恢复向量的问题X从嘈杂的观察一个X. 这里又是最受欢迎的模型这是标准高斯噪声。

## 统计代写|统计推断代写Statistical inference代考|Compressed Sensing via ℓ1 minimization: Motivation

\min {z}\left{|z|{0}: A z=y\right},\min {z}\left{|z|{0}: A z=y\right},

\min {z}\left{|z|{0}:|A zy| \leq \epsilon\right} 。\min {z}\left{|z|{0}:|A zy| \leq \epsilon\right} 。

|一个¯在|≥C−1|在|2

## 有限元方法代写

tatistics-lab作为专业的留学生服务机构，多年来已为美国、英国、加拿大、澳洲等留学热门地的学生提供专业的学术服务，包括但不限于Essay代写，Assignment代写，Dissertation代写，Report代写，小组作业代写，Proposal代写，Paper代写，Presentation代写，计算机作业代写，论文修改和润色，网课代做，exam代考等等。写作范围涵盖高中，本科，研究生等海外留学全阶段，辐射金融，经济学，会计学，审计学，管理学等全球99%专业科目。写作团队既有专业英语母语作者，也有海外名校硕博留学生，每位写作老师都拥有过硬的语言能力，专业的学科背景和学术写作经验。我们承诺100%原创，100%专业，100%准时，100%满意。

## MATLAB代写

MATLAB 是一种用于技术计算的高性能语言。它将计算、可视化和编程集成在一个易于使用的环境中，其中问题和解决方案以熟悉的数学符号表示。典型用途包括：数学和计算算法开发建模、仿真和原型制作数据分析、探索和可视化科学和工程图形应用程序开发，包括图形用户界面构建MATLAB 是一个交互式系统，其基本数据元素是一个不需要维度的数组。这使您可以解决许多技术计算问题，尤其是那些具有矩阵和向量公式的问题，而只需用 C 或 Fortran 等标量非交互式语言编写程序所需的时间的一小部分。MATLAB 名称代表矩阵实验室。MATLAB 最初的编写目的是提供对由 LINPACK 和 EISPACK 项目开发的矩阵软件的轻松访问，这两个项目共同代表了矩阵计算软件的最新技术。MATLAB 经过多年的发展，得到了许多用户的投入。在大学环境中，它是数学、工程和科学入门和高级课程的标准教学工具。在工业领域，MATLAB 是高效研究、开发和分析的首选工具。MATLAB 具有一系列称为工具箱的特定于应用程序的解决方案。对于大多数 MATLAB 用户来说非常重要，工具箱允许您学习应用专业技术。工具箱是 MATLAB 函数（M 文件）的综合集合，可扩展 MATLAB 环境以解决特定类别的问题。可用工具箱的领域包括信号处理、控制系统、神经网络、模糊逻辑、小波、仿真等。