The aim of the project is to investigate current computational approaches for algorithmic stability, convergence and consistency of representative methods of dynamic machine learning (e.g. Recurrent (Convolutional) Neural Networks, Dynamic Graph Convolutional Networks, etc.). In the first phase, well-known dynamic ML methods will be categorized according to certain criteria (including complexity, performance and stability) and representative models will be selected. In the second phase, these models will be used to assess various metrics for stability, convergence and consistency with regard to their effectiveness and applicability. At the same time, qualitative and quantitative criteria for the susceptibility of the models to errors are to be developed. In the last section, recommendations for adapting the models are derived and applied. The recommendations will be tested on existing metrics of robustness and reliability.

See II.1.4)