Deep learning based on multilayer neural networks have recently become a state-of-the-art method in machine learning for classification. They may be used for significant economic and industrial applications (e.g., credit scoring, monitoring of critical production processes or the safety of computer networks). The applications of those methods heavily depend on the homogeneity of the data over time. Therefore, developing methods for checking these assumptions are essential, but do not yet exist for such complex networks. The goal of this project is to develop tests for the presence of changes in time for multilayer neural networks based on previous work on single-layer networks (Kirch and Tadjuidje, 2012, 2014) and parameter estimation for multilayer networks (Bauer and Kohler, 2017).