July 17, 2023 - by Simone Ulmer
“The PASC Conference is purely about science — it is the only conference in the field of high-performance computing (HPC) that brings together researchers and vendors without a showroom,” noted Erik Lindahl, professor of biophysics at Stockholm University and KTH, in an interview during PASC23. Over 160 researchers presented their work at the Platform for Advanced Scientific Computing (PASC) conference in the form of 44 topically focused minisymposia, triggering a lively exchange between the more than 400 conference participants who converged in Davos from June 26–28, 2023. The presentations centered around this year's theme: "Computing Across Scales, Domains, and Communities".
Lindahl and Associate Director of the Swiss National Supercomputing Centre (CSCS) Maria Grazia Giuffreda explored the impact of the minisymposia by conducting more than 40 interviews with the chairs of the minisymposia to find out whether this form of scientific and interdisciplinary dialogue influenced or perhaps even advanced their research. Was it worth the effort for the researchers? Did helpful and further discussions arise, and did new and interdisciplinary collaborations emerge, as intended by the organisers?
“The future is bright.”
The interviews revealed that the chairs were satisfied and pleased that the minisymposia enabled both exciting discussions on their research area and new contacts for possible research collaborations. You can watch the full interviews on the PASC23 Youtube channel. The feedback from the researchers was entirely in keeping with the positive mood of the conference. In an interview with Lindahl, ETH Professor and CSCS Director Thomas Schulthess, who initiated both PASC and its projects as well as the PASC Conference, also expressed his satisfaction about the evolution that the event has taken over the past decade.
Schulthess pointed to various projects in earth sciences, especially weather and climate, but also astrophysics (SKA) as examples attesting to the success of PASC. “I think the future is bright. Given that we now know, together with the community in Switzerland, how to address all the various architectural problems on the computing side, we can finally tackle the next frontier, which, I believe, is scientific workflows including non-trivial interaction with scientific data.” When asked what he would change, Schulthess said: “The conference is a great success — don’t change a success.” The only thing he would like to see, he added, would be even more interactions between the domains.
"Don't say sorry, just improve!"
The welcome message was delivered by the local hosts and this year’s conference Chairs, ETH Professor Eleni Chatzi and Senior Scientist at Sandia National Laboratory, USA, Mike Heroux. Their introductions were followed by the first keynote, given by Lorena A. Barba, professor of mechanical and aerospace engineering at George Washington University in Washington, DC. Barba discussed the topic “Anti-patterns of Scientific Machine Learning to Fool the Masses: A Call for Open Science”. According to Barba, an anti-pattern is a frequently occurring pattern that is ineffective and risks being counterproductive. Barba showed a series of example how anti-patterns while using ML methods and publishing results obstruct the exploitation of the potential of machine learning. “Don't say sorry, just improve!", Barba advised a participant who tried to explain why academic work in ML sometimes faces challenges and falls short. According to Barba, the path of improvement is establishing a culture and practices that lead to a transparent scientific process for researching and publishing.
The conference program included five keynotes in all, as well as one interdisciplinary dialogue, a panel discussion, the 44 minisymposia, and two exciting Flash Poster sessions. During the poster sessions, 57 young emerging researchers had the opportunity to pitch the topic of their poster within 30 seconds, thus attracting visitors and convincing them to vote for it as the best conference poster. That prize was ultimately awarded to two presenters: Lidia Caros from Peter Vincent's research group at Imperial College London, who optimised non-conventional airfoils for Martian rotorcrafts using numerical simulations on CSCS’s “Piz Daint”; and Ivan Pribec from the Leibniz Supercomputing Centre in Germany, who addressed the never-ending dispute between Fortran and C++, showing the interoperability of the two languages.
In addition, there were numerous ACM Paper sessions, at which the authors of 26 papers presented their research results. The best paper award at PASC23 went to Thorsten Kurth and Karthik Kashinath from NVIDIA Switzerland for their publication “FourCastNet: Accelerating Global High-Resolution Weather Forecasting Using Adaptive Fourier Neural Operators.” Paper program committee Co-Chair Axel Huebl from Lawrence Berkeley Laboratory pointed out that their publication showed that using machine learning in weather simulations allows going to large scales and achieving extremely fast speed-ups while still maintaining predictive quality. Kurth also presented his results in his Keynote speech on the last conference day.
The interdisciplinary dialogue took place in the early morning of the second day. The goal of the interdisciplinary dialogue is to give researchers insight into each other’s research disciplines and also to inspire their own research work. This year, ETH Professor Marina Krstic Marinkovic interviewed Professor Jens Honoré Walther about: “Flow Simulations Across Scales: From Atoms to Vortices”. Walther teaches computational fluid dynamics at the Technical University of Denmark, and his research spans length scales from nano to kilometers — from numerical studies of capillarity in nanofluidic systems to bluff body aerodynamics of suspension bridges. As a physicist, Marinkovic explores the interplay between lattice quantum chromodynamic (QCD) and lattice quantum electrodynamics (QED) analytically and numerically, through algorithms optimized to run efficiently on HPC infrastructures.
The conference program also included a Public Lecture from Argonne National Laboratory’s Rick Stevens, who is a professor of computer science at the University of Chicago. The title of his presentation was: “Exascale and then what … A view of Post-Exascale Computational Science and the Emerging Mix of HPC, AI and Quantum”. In his 70-minute talk, he explored the US-American history of HPC exascale computing and how numerous workshops from 2007 onward ultimately resulted in science cases for exascale in 2012, based on science opportunities and social needs. The science cases included, for example, nuclear energy, climate geoengineering, and materials by design.
Stevens further explained how domain-specific general AI foundation models can be used for handling most of the use cases described in the latest report from Argonne National Laboratory on AI for Science, Energy, and Security. However, Stevens was rather skeptical with respect to quantum computing. First of all, he said, it will not be the magic wand that solves all problems but must always be embedded in a classical compute environment. Second, it is at this point unclear which will be the practical “killer applications” for quantum computing, although some potential candidates have already been identified. Finally, the cost of these systems is yet unclear. If one use last year’s price for a qubit and extrapolate to the estimated needs of a true application, a quantum computer using the price development of classical supercomputers over the last 20 years probably won’t be cost-competitive with a future zettascale system.
Nothing replaces imagination
The final day of the conference opened with keynotes from the awardees. Petros Koumoutsakos, Professor for Computing in Science and Engineering and the Area Chair of Applied Mathematics at Harvard University, was awarded the PRACE HPC excellence award. Over twenty-five years, Koumoutsakos actively shaped and advanced computational research in Europe as an ETH professor and Chair for Computational Science at ETH. In his talk titled “AI, Computing and Thinking: Algorithmic Alloys for Advancing Scientific Discovery”, Koumotsakos gave an overview of his research over the years and how he now combines multi-agent reinforcement learning and HPC in simulating multi-scale systems for modelling control of biological and turbulent flows. He also showed the current limitations of machine learning, for example, by trying to recreate the classical simulated evolution of dynamic flow over time using neural networks. What looked good at first failed increasingly over time. “Neuronal networks cannot replace numerical methods for the long run”, he concluded.
Koumoutsakos is convinced that computing transforms our thinking for solving complex problems and fuels the AI revolution that is changing our world. He believes that we are at the dawn of a new era in science that would benefit from forming “alloys” of AI, computing and thinking. But: “There will not be a day where machine learning or other things will replace imagination; however, they will complement.”
This year's PRACE Ada Lovelace Award for HPC was awarded to Computer Scientist Sarah Neuwirth from Goethe University in Frankfurt. Neuwirths’ work forms an important interface between users, HPC applications, and their performance on the computers. With the increasing complexity of computer architectures, as well as the use of ML and AI, special attention, like her work, is needed so that the codes used can optimally exploit the theoretical performance of the computer for their computation.
The conference program closed with panelists from different institutions and with different backgrounds discussing the theme “Better Together: How Can We Enhance Inter-Community Collaboration?” (For more information, see the full video).
After PASC23 is before PASC24
PASC24 will take place from June 3–5, 2024, at ETH Zurich Hönggerberg Campus. The conference chairs next year will be the Director for the Computational Sciences and Engineering Division at Oak Ridge National Laboratory, Katherine Evans, and Professor for Computational and Applied Mathematics at ETH Zurich, Siddhartha Mishra. The planned theme will be “Synthesizing Applications Through Learning and Computing”. The conference will focus on the grand challenges of combining physics-based simulations with novel machine learning and AI-based methods to address interdisciplinary problems in science. This is entirely in keeping with the spirit of the “inventor” of the PASC Conference, Schulthess, who would like to see even more interaction between the domains.
More information:
The Platform for Advanced Scientific Computing (PASC) Conference is an interdisciplinary conference in HPC that brings together domain science, applied mathematics, and computer science – where computer science is focused on enabling the realization of scientific computation.
The PASC Conference provides three days of stimulating and thought-provoking technical sessions, including keynote presentations, minisymposia, peer-reviewed papers, panels, and poster sessions. The conference is co-sponsored by ACM SIGHPC, and full papers are published in the ACM Digital Library.