Dutch collaboration wins HPC Innovation Excellence Award for the first time
A Dutch collaboration, including the SURF Open Innovation Lab and Leiden Observatory, has won Hyperion Research's HPC Innovation Excellence Award. This is the first time that a Dutch team has won the award. The team received the award for improving large-scale numerical simulations with deep learning.
The HPC Innovation Excellence Awards distinguish outstanding achievements of users of high performance computing (HPC), including simulation, AI and quantum computing.
Proud and honoured
Axel Berg, Innovation Manager Lab at SURF: ‘We are proud and honoured that our team received this award. We set up and executed a successful collaboration with a number of academic groups in various scientific domains, in a new and unexplored approach of applying deep learning techniques.’
Professor Simon Portegies Zwart explains how this collaboration came about: ‘It started with an initiative of SURF colleagues, who were interested in how we can use large computers more efficiently. They had the idea that machine learning might be a great opportunity to make that happen. Coincidentally, we in Leiden were working on exactly the same question. After a phone call from Axel it was immediately clear that we actually had an identical idea, but from a different point of view.’
SURF consultant Caspar van Leeuwen: ‘Together with the other participants, we saw that machine learning might offer an opportunity to make numerical simulations more efficient – and therefore faster. This shared vision and interest has therefore brought us together.’
Portegies Zwart: ‘I think we have won this prize because we use the computer and artificial intelligence in a completely different way than others. Our approach is truly unique, not only from a scientific-technical point of view, but also from a scientific-philosophical point of view.’
Applying Deep Learning
The guidelines and approaches developed by the team are generic and can be useful for many research communities and companies. Applications include improving large-scale numerical simulations, speeding up simulations, or increasing problem and data size. In addition, the experiments carried out in the various experiments have led to real – and in some cases groundbreaking – improvements.
It is not yet entirely clear what the exact consequences will be for science. ‘We have shown that this approach really adds something to the way we view our numerical problems,’ says Portegies Zwart. He calls the research an experiment to see what would happen. ‘I think that this paved the way for the real scientific breakthroughs.’
Van Leeuwen adds: ‘We have shown that machine learning can be an extra piece of “tool” in the toolbox we have for solving numerical problems. As Simon said, what we can ultimately achieve in terms of scientific results with that extra tool will have to be learnt over time.’
The project was a collaboration between the SURF Open Innovation Lab team and academic groups from Leiden Observatory, Leiden University; Meteorology and Air Quality Group, Wageningen University and Research; Institute for Mathematics, Astro- and Particle Physics IMAPP, Radboud University; and Bijvoet Centre for Biomolecular Research, Faculty of Science - Chemistry, Utrecht University.
This news article is based on a previous article by SURF