1: \begin{abstract}

2: We study estimation of a multivariate function 

3: $f:{\bf R}^d \to {\bf R}$ 

4: when the observations are available from function $Af$, where

5: $A$ is a known linear operator. Both the Gaussian white noise model

6: and density estimation are studied. We define an $L_2$ empirical

7: risk functional, which is used to define an $\delta$-net minimizer

8: and a dense empirical risk minimizer. Upper bounds for the mean

9: integrated squared error of the estimators are given. The upper

10: bounds show how the difficulty of the estimation depends on the

11: operator through the norm of the adjoint of the inverse of the

12: operator, and on the underlying function class through the entropy

13: of the class. Corresponding lower bounds are also derived. As

14: examples we consider convolution operators and the Radon transform.

15: In these examples the estimators achieve the optimal rates of

16: convergence.

17: Furthermore, a new type of oracle inequality is given for

18: inverse problems in additive models.

19: \end{abstract}

20: