1: \begin{abstract}

2: In the present paper, we constructed an estimator of a delta contaminated mixing density function $g(\lam)$ 

3: of the intensity $\lambda$ of the Poisson distribution.

4: The estimator is based on an expansion of the continuous portion $g_0(\lambda)$ of the unknown pdf over an overcomplete dictionary

5: with the recovery of the coefficients  obtained as   solution of an optimization problem with Lasso penalty. In order to  

6: apply Lasso technique in the, so called, prediction setting where it requires virtually no assumptions on dictionary 

7: and, moreover, to ensure fast convergence of Lasso estimator, we use a novel formulation of the optimization problem

8: based on inversion of the dictionary elements. 

9: The total estimator of the  delta contaminated mixing pdf is obtained using a two-stage iterative procedure.

10: 

11: 

12: We formulate conditions on the dictionary and the unknown mixing density that yield a sharp oracle inequality 

13: for the norm of the difference between $g_0 (\lambda)$ and its estimator and, thus, obtain a   smaller error  than

14: in a minimax setting. Numerical simulations and comparisons with the  Laguerre functions based estimator 

15: recently constructed by Comte and Genon-Catalot  (2015) also show advantages of our procedure. 

16: At last, we apply the technique  developed in the paper to estimation of 

17: a delta contaminated mixing density of the Poisson intensity of the Saturn's rings data.\\

18: %

19: \vspace{3mm} {\bf  Keywords}: {Mixing density, Poisson distribution, empirical Bayes, Lasso penalty }\\

20: %

21: \vspace{3mm}{\bf AMS (2000) Subject Classification}: {Primary  62G07, 62C12. Secondary 62P35}

22: 

23: 

24: \end{abstract}

25: