De-sparsified lasso contributes to construct confidence intervals and statistical tests for single or low-dimensional components of a large parameter vector in high-dimensional model.^[1]

This article has multiple issues. Please help improve it or discuss these issues on the talk page. (Learn how and when to remove these template messages) This article is an orphan, as no other articles link to it. Please introduce links to this page from related articles; try the Find link tool for suggestions. (July 2016) This article needs attention from an expert in statistics. The specific problem is: needs a better lead and overview, more general context and categorization. (June 2015) This article relies largely or entirely on a single source. (October 2019)

${\displaystyle Y=X\beta ^{0}+\epsilon }$ with ${\displaystyle n\times p}$ design matrix ${\displaystyle X=:[X_{1},...,X_{p}]}$ (${\displaystyle n\times p}$ vectors ${\displaystyle X_{j}}$), ${\displaystyle \epsilon \sim N_{n}(0,\sigma _{\epsilon }^{2}I)}$ independent of ${\displaystyle X}$ and unknown regression ${\displaystyle p\times 1}$ vector ${\displaystyle \beta ^{0}}$.

The usual method to find the parameter is by Lasso: ${\displaystyle {\hat {\beta }}^{n}(\lambda )={\underset {\beta \in \mathbb {R} ^{p}}{argmin}}\ {\frac {1}{2n}}\left\|Y-X\beta \right\|_{2}^{2}+\lambda \left\|\beta \right\|_{1}}$

The de-sparsified lasso is a method modified from the Lasso estimator which fulfills the Karush–Kuhn–Tucker conditions^[2] is as follows:

${\displaystyle {\hat {\beta }}^{n}(\lambda ,M)={\hat {\beta }}^{n}(\lambda )+{\frac {1}{n}}MX^{T}(Y-X{\hat {\beta }}^{n}(\lambda ))}$

where ${\displaystyle M\in R^{p\times p}}$ is an arbitrary matrix. The matrix ${\displaystyle M}$ is generated using a surrogate inverse covariance matrix.

Desparsifying ${\displaystyle l_{1}}$-norm penalized estimators and corresponding theory can also be applied to models with convex loss functions such as generalized linear models.

Consider the following ${\displaystyle 1\times p}$vectors of covariables ${\displaystyle x_{i}\in \chi \subset R^{p}}$ and univariate responses ${\displaystyle y_{i}\in Y\subset R}$ for ${\displaystyle i=1,...,n}$

we have a loss function ${\displaystyle \rho _{\beta }(y,x)=\rho (y,x\beta )(\beta \in R^{p})}$ which is assumed to be strictly convex function in ${\displaystyle \beta \in R^{p}}$

The ${\displaystyle l_{1}}$-norm regularized estimator is ${\displaystyle {\hat {\beta }}={\underset {\beta }{argmin}}(P_{n}\rho _{\beta }+\lambda \left\|\beta \right\|_{1})}$

Similarly, the Lasso for node wise regression with matrix input is defined as follows: Denote by ${\displaystyle {\hat {\Sigma }}}$ a matrix which we want to approximately invert using nodewise lasso.

The de-sparsified ${\displaystyle l_{1}}$-norm regularized estimator is as follows: ${\displaystyle {\hat {\gamma _{j}}}:={\underset {\gamma \in R^{p-1}}{argmin}}({\hat {\Sigma }}_{j,j}-2{\hat {\Sigma }}_{j,/j}\gamma +\gamma ^{T}{\hat {\Sigma }}_{/j,/j}\gamma +2\lambda _{j}\left\|\gamma \right\|_{1}}$

where ${\displaystyle {\hat {\Sigma }}_{j,/j}}$ denotes the ${\displaystyle j}$th row of ${\displaystyle {\hat {\Sigma }}}$ without the diagonal element ${\displaystyle (j,j)}$, and ${\displaystyle {\hat {\Sigma }}_{/j,/j}}$ is the sub matrix without the ${\displaystyle j}$th row and ${\displaystyle j}$th column.