Materials with a complex microstructure have several special features manifesting at different length scales and causing various behaviour patterns. This motivates the development of numerous numerical methods for their investigation. On the one hand, the multiscale approaches summarize the information on the material behaviour at lower length scales to predict effective structural behaviour, following different assumptions such as volume averaging, statistical averaging, or energy balance expressed through the Hill-Mandel upscaling condition. On the other hand, the complexity of the material behaviour under different loading cases pushes the development of ML-based approaches. Here, classical methods characterized by the evolution of internal state variables are replaced by Neural Networks (NN) that are able to memorize. Typically, the focus is placed on RNN, LSTM, and GRU, but the alternative NNs, such as Physics-Informed NNs architectures, are also in use. The latter are especially advantageous since they address the question of a limited amount of informative data, the question of physical admissibility, the need to explicitly deal with uncertainties, and the need to provide explainable and interpretable inferences. The minisymposium covers novel numerical aspects in both kinds of approaches, multiscale and ML-based ones, as well as their contribution to a better understanding and improvement of the material behaviour of structural materials, especially 3D printed materials.