This project develops ML-enhanced techniques for improving the processing of data from ultracold atom experiments. We optimize the analysis of fluorescence and absorption images in optical lattice experiments where site-resolved atom numbers and phases are to be reconstructed. Furthermore, we develop methods for the reconstruction of (sub)system states that optimally exploit prior knowledge in the form of physical constraints or ML inspired variational ansatz functions – a dimensional reduction task. Finally, we develop adaptive measurement strategies for choosing the optimal measurement settings based on previous observations on the system. Here ML methods allow us to overcome the prohibitive numerical complexity of traditional methods.