Since the early 2000s, density functional theory (DFT) calculations have become widely adopted, focusing on idealized systems with simple surfaces and small reaction networks. However, real-world systems involve complex morphology, time-dependent atomistics dynamics, and numerous reactions, and intermediates. In this context, machine learning has had a significant impact on material development and understanding, offering flexible predictions of material properties. Machine learning can unravel the complex, nonlinear correlations between atomic structures and catalytic activity, enabling predictions for various materials phenomena without time-consuming and costly experimental tests or DFT calculations. This seminar will discuss how machine learning can be applied to study complex systems, such as understanding the hydrogen evolution reaction mechanisms of jagged Pt nanowires, solid oxygen conductor as well as the synthesizability of hypothetical crystals. It will also demonstrate how machine learning models can be integrated into STEM to enhance material analysis at the atomic level.