State-of-the-art data-driven inference methods and magnetic multiscale modelling are combined to systematically study material inhomogeneities and characterize their role in the hysteresis behaviour. A large amount of micromagnetic and atomistic spin dynamics simulations with various inhomogeneities will be produced to train a machine learning model (A01, A05, A06, A07, B06, B13), which can automatically detect, and characterize different inhomogeneities and their influence on the hysteresis behaviour (B01, B11). Exploiting this analysis framework, experimental data will be analysed.