Session

DescriptionHigh-performance computing (HPC) has a long history of driving scientific discovery through advances in hardware and numerical algorithms, but the adoption of artificial intelligence (AI) and machine learning (ML) is transforming this landscape. By integrating traditional simulations with AI/ML training and inference tasks into complex workflows, computational scientists are unlocking new HPC applications, from AI-driven design and optimization to online model fine-tuning and learning of dynamical systems, and revolutionizing how we tackle many of the UN’s sustainable development goals. However, building and efficiently deploying large-scale coupled workflows on HPC systems still poses significant software and hardware challenges, including managing massive datasets on distributed systems, making efficient use of the interconnect and local memory to avoid I/O bottlenecks, and ensuring reproducibility and provenance. In this minisymposium, speakers from leading hardware vendors, HPC centers, and universities share the latest software innovations, new learning methodologies developed, and successful practices adopted to address the issues faced by coupled simulation and AI workflows on modern HPC systems. Through applications in fields such as drug discovery and climate modeling, the talks will discuss lessons learned and the remaining challenges in adopting large-scale coupled workflows for scientific discovery in the exascale era of supercomputing.