Hardware-Informed Applications, Application-Informed Hardware

MACHINE learning and Scientific Computing stand to gain significantly from Quantum Computing (QC), which holds promise to deliver new tools for solving complex and computationally demanding problems. Hardware-informed approaches, which attempt to lower the bar in resources required for the utilization of quantum paradigms, could accelerate such development. Simultaneously, as traditional workflows routinely employed in a number of applications evolve, so too do the opportunities for integrating quantum algorithms in the pipeline and unlock classical limitations. By developing the understanding for distinct advantages of quantum devices, new methods can emerge to enhance predictive accuracy, accelerate simulations, and uncover novel insights in scientific and industrial applications. However, limited resources offered by first generations of quantum devices typically necessitate of classical co-processing power, and require tailored software-hardware approach, often referred as ’co-design’. Furthermore, the nature of quantum computing and machine learning in science & technology applications increasingly call for methods that are physics-informed. This review explores the intersection of quantum computing and machine learning, with a particular emphasis on scientific machine learning (SciML). Unlike conventional machine learning, SciML is deeply rooted in formal and domain-specific constraints, making it an ideal candidate for quantum-enhanced approaches. Crucially, we argument how effective quantum algorithms need to both be aligned with the structures and requirements of targeted scientific problems, whilst carefully understanding the interplay between hardware and problem formulation. A key focus of this review is the role of neutral atomic quantum processors in advancing SciML. These platforms offer unique computational capabilities that remain under-explored compared to other quantum platforms. With their ability to naturally encode spatial and graph-based problems, neutral atom systems provide a promising avenue for tackling optimization and machine learning tasks that are particularly relevant in scientific computing. By highlighting in few examples the advantages and open questions surrounding these processors, we aim to present a forward-looking perspective on how quantum computing can reshape computational approaches in science and engineering.