1: \begin{abstract}

2: Neural networks have become ubiquitous tools for solving signal and

3: image processing problems, and they often outperform standard

4: approaches. Nevertheless, training neural networks is a

5: challenging task in many applications. The prevalent training

6: procedure consists of minimizing highly non-convex objectives

7: based on data sets of huge dimension. In this context, current

8: methodologies are not guaranteed to produce global solutions. We

9: present an alternative approach which foregoes the optimization

10: framework and adopts a variational inequality formalism. The

11: associated algorithm guarantees convergence of the iterates to a

12: true solution of the variational inequality and it possesses an

13: efficient block-iterative structure. A numerical application is

14: presented.

15: \end{abstract}

16: