Conception, design and operation of complex systems typically entail solving parametric and possibly time-dependent differential equations. This is particularly challenging in several contexts, such as inverse problems, uncertainty quantification, data assimilation, real-time monitoring, control, and optimisation. Efficient surrogate, reduced-order, and multi-fidelity modelling techniques are among the key enabling technologies to make such problems affordable. Therefore, this minisymposium aims to offer a platform for discussing methodological developments and practical challenges of frontier scientific machine learning techniques. This comprises latest advances on physics-informed and data-driven methodologies such as proper orthogonal and proper generalised decompositions, collocation methods (radial basis functions, Gaussian processes, sparse grids, …), polynomial chaos expansion, dynamic mode decomposition, low-rank approximation, tensor-train decomposition, and neural networks. Contributions tackling the use of the aforementioned methods in state-of-the-art applications and challenging scenarios such as (but not limited to) multi-physics, multi-scale, and coupled problems in industry, engineering, environmental and sustainability sciences are particularly welcome.