BACKGROUND: To say data is revolutionising the medical sector would be a vast understatement. The amount of medical data available today is unprecedented and has the potential to enable to date unseen forms of healthcare. To process this huge amount of data, an equally huge amount of computing power is required, which cannot be provided by regular desktop computers. These areas can be (and already are) supported by High-Performance-Computing (HPC), High-Performance Data Analytics (HPDA), and AI (together “HPC+”). OBJECTIVE: This overview article aims to show state-of-the-art examples of studies supported by the National Competence Centres (NCCs) in HPC+ within the EuroCC project, employing HPC, HPDA and AI for medical applications. METHOD: The included studies on different applications of HPC in the medical sector were sourced from the National Competence Centres in HPC and compiled into an overview article. Methods include the application of HPC+ for medical image processing, high-performance medical and pharmaceutical data analytics, an application for pediatric dosimetry, and a cloud-based HPC platform to support systemic pulmonary shunting procedures. RESULTS: This article showcases state-of-the-art applications and large-scale data analytics in the medical sector employing HPC+ within surgery, medical image processing in diagnostics, nutritional support of patients in hospitals, treating congenital heart diseases in children, and within basic research. CONCLUSION: HPC+ support scientific fields from research to industrial applications in the medical area, enabling researchers to run faster and more complex calculations, simulations and data analyses for the direct benefit of patients, doctors, clinicians and as an accelerator for medical research.

Scientific communities are increasingly adopting machine learning and deep learning models in their applications to accelerate scientific insights. High performance computing systems are pushing the frontiers of performance with a rich diversity of hardware resources and massive scale-out capabilities. There is a critical need to understand fair and effective benchmarking of machine learning applications that are representative of real-world scientific use cases. MLPerf is a community-driven standard to benchmark machine learning workloads, focusing on end-to-end performance metrics. In this paper, we introduce MLPerf HPC, a benchmark suite of large-scale scientific machine learning training applications driven by the MLCommons Association. We present the results from the first submission round, including a diverse set of some of the world's largest HPC systems. We develop a systematic framework for their joint analysis and compare them in terms of data staging, algorithmic convergence, and compute performance. As a result, we gain a quantitative understanding of optimizations on different subsystems such as staging and on-node loading of data, compute-unit utilization, and communication scheduling, enabling overall $>10 \times$ (end-to-end) performance improvements through system scaling. Notably, our analysis shows a scale-dependent interplay between the dataset size, a system's memory hierarchy, and training convergence that underlines the importance of near-compute storage. To overcome the data-parallel scalability challenge at large batch sizes, we discuss specific learning techniques and hybrid data-and-model parallelism that are effective on large systems. We conclude by characterizing each benchmark with respect to low-level memory, I/O, and network behavior to parameterize extended roofline performance models in future rounds.

In the context of network information theory, one often needs a multiparty probability distribution to be typical in several ways simultaneously. When considering quantum states instead of classical ones, it is in general difficult to prove the existence of a state that is jointly typical. Such a difficulty was recently emphasized and conjectures on the existence of such states were formulated. In this paper, we consider a one-shot multiparty typicality conjecture. The question can then be stated easily: is it possible to smooth the largest eigenvalues of all the marginals of a multipartite state ρ simultaneously while staying close to ρ? We prove the answer is yes whenever the marginals of the state commute. In the general quantum case, we prove that simultaneous smoothing is possible if the number of parties is two or more generally if the marginals to optimize satisfy some non-overlap property.