Sparse Poisson regression with penalized weighted score function (Q2323944)

This paper proposes an estimation approach for the high-dimensional heteroscedastic Poisson regression model based on the optimization of a new \(\ell _1\) -penalized weighted score function. After adding a weight to the score function, the penalty coefficient does not depend on the variance of the Poisson noise. It is shown that the new estimator is consistent, and the convergence rate to the true parameter is explicited. Comparisons are developed on the basis of simulation and a real application to image reconstruction.