1: \begin{abstract}

2: 	Neural networks functions are supposed to be able to encode the desired solution of an inverse problem 

3: 	very efficiently.  

4: 	In this paper, we consider the problem of solving linear inverse problems with neural network coders. 

5: 	First we establish some correspondences of this formulation with existing concepts in regularization theory, 

6: 	in particular with state space regularization, operator decomposition and 

7: 	iterative regularization methods. A Gauss-Newton's method is suitable for solving encoded linear inverse problems, 

8: 	which is supported by a local convergence result. 

9: 	The convergence studies, however, are not complete, and are based on a conjecture on linear independence of 

10: 	activation functions and its derivatives.

11: 	\bigskip\par\noindent

12: 	This paper is dedicated to Heinz Engl's 70th birthday.

13: \end{abstract}

14: