1: \begin{abstract}

2: In \cite{iterative}, a least square regression estimation procedure was proposed: first, we condiser

3: a family of functions $f_{k}$ and study the properties of an estimator in every unidimensionnal model

4: $\{\alpha f_{k},\alpha\in\mathds{R}\}$;

5: we then show how to aggregate these estimators. The purpose of this paper is to extend this method to the case

6: of density estimation.

7: We first give a general overview of the method, adapted to the density estimation problem. We then show that this leads to adaptative estimators,

8: that means that the estimator reaches the best possible rate of convergence (up to a $\log$ factor).

9: Finally we show some ways to improve and generalize the method.

10: \end{abstract}

11: