1: \begin{abstract}

2:     Neural networks have revolutionized the field of data science, yielding remarkable solutions in a data-driven manner. 

3:     For instance, in the field of mathematical imaging, they have surpassed traditional methods based on convex regularization.

4:     However, a fundamental theory

5:     supporting the practical applications is still in the early stages of development. 

6:     We take a fresh look at neural networks and examine them via nonlinear eigenvalue analysis.

7:     The field of nonlinear spectral theory is still emerging, providing insights about nonlinear operators and systems. 

8:     In this paper we view a neural network as a complex nonlinear operator and attempt to find its nonlinear eigenvectors. 

9:     We first discuss the existence of such eigenvectors and analyze the kernel of $\relu$ networks.

10:     Then we study a nonlinear power method for generic nonlinear operators. 

11:     For proximal operators associated to absolutely one-homogeneous convex regularization functionals, we can prove convergence of the method to an eigenvector of the proximal operator.

12:     This motivates us to apply a nonlinear method to networks which are trained to act similarly as a proximal operator.

13:     In order to take the non-homogeneity of neural networks into account we define a modified version of the power method.

14:     \\

15:     We perform extensive experiments \revise{for different proximal operators and} on various shallow and deep neural networks designed for image denoising.

16:     \revise{Proximal eigenvectors will be used for geometric analysis of graphs, as clustering or the computation of distance functions.}

17:     For simple neural nets, we observe the influence of training data on the eigenvectors.

18:     For state-of-the-art denoising networks, we show that eigenvectors can be interpreted as (un)stable modes of the network, when contaminated with noise or other degradations.

19: \end{abstract}

20: