Despite the ongoing pandemic that already forced the organizers to postpone PASC20 at University of Geneva, the PASC21 conference took place, albeit in a virtual format, from July 05 to 09, 2021. In-person attendance on the beautiful shores of Lake Geneva was still not possible. Nevertheless, the conference turned out to be a perfectly organized event, mostly virtual, with few hybrid exceptions for the benefit of the virtual audience composed of 470 scientists from all over the world. Maria Grazia Giuffreda, associate director at CSCS, looks back on an exciting event.
August 26, 2021 – by Maria Grazia Giuffreda, CSCS
The theme of the conference “New Challenges, New Computing Paradigms” reflects the new exciting era we are facing, mentioning only Exascale Computing, Quantum Computing, Artificial Intelligence, and Machine Learning. All of them require new ideas and significant changes in hardware design and computing paradigms to tap the full potential of compute and storage power.
The chairs of the conference welcomed the participants in an unusual format. Bastien Chopard, professor in the Computer Science Department at University of Geneva spoke from the media room at the host University of Geneva, while Sinéad Ryan, chair of Theoretical High Energy Physics at Trinity College Dublin, connected remotely from her office. Together they presented the rich program of this year’s conference, which they both contributed to put together. There were two keynotes and one public lecture, one interdisciplinary dialogue and one panel discussion, a poster session, as well as 59 minisymposia.
The Golden era of Computer Architectures
Viv Kendon, reader in the Department of Physics at Durham University opened the conference with a keynote lecture on the prospects and challenges of quantum computing, underlining the amazing activities around it but also warning of the current approach taken by many: It is somehow worrisome that there are more quantum software startups than existing hardware, concentrating on building python interfaces and libraries even though it is impossible to test such software for as long as there is no hardware around. Jeronimo Castrillon, chair for Compiler Construction at TU Dresden was the keynote speaker to close the scientific program of this year’s conference with an enthusiastic talk about domain specific languages for heterogeneous and emerging computer architectures. The driving force of his research is to ensure that the golden era of computer architecture is of benefit to everybody and not just a playground for Ninja programmers.
The public lecture was held by Dalia Conde, professor in the department of Biology at the University of Southern Denmark and director of science at Species360. She focused on the use of machine learning, big data, and exascale computing in addressing biodiversity loss, one of the threats that our society is facing. Many organizations are concerned with the current extinction trend and for decision making they need to rely on data. Thus, one of the ideas is to try to explore and map all this data and extract knowledge. Conde concluded her talk with a strong message: Current extinction is not caused by unpreventable natural disaster such as meteorite impact that made dinosaurs disappear. It is driven by humans, by all of us, and we are in a position of reverting the trend of extinction if we start acting right now!
Interdisciplinary dialogue: “I love Biology. Why don’t biologists love theory?”
The interdisciplinary dialogue saw two eminent scientists, Peter Coveney from University College London and Erik Lindahl from Stockholm University, debating about Physics, Biology and Computer science. They kept the audience enthralled with a lively discussion about why biologists seem not overly excited about theory and dispelled an old myth that mathematics being the very difficult discipline and biology the very easy one. It actually seems to be the other way around because in biology there are no exact equations that let us describe and understand very complex mechanisms and structures. This dialogue raised the question whether scientists might be stuck in an old formulation of problems that prevents them to make use of the already enormous amount of available computational resources, as they are applying “old” methods to “new” challenges. Or whether scientists are becoming obsessed with extremely complicated problems rather than focus on those much simpler that might have a big impact on society and would also be appreciated by the masses. This is food for thought for present and future generations of scientists. Once again it became clear how increasingly important it is to overcome artificial boundaries between disciplines and learn how to work together.
AI is a powerful tool but not a replacement for human ingenuity
The Panel discussion was an unusual event as Maria Girone, CTO of CERN OpenLab and François Fleuret, professor in the department of Computer Science at the University of Geneva, were sitting in the media room at University of Geneva together with the chair Bastien Chopard, whereas the three other panelists, Trilce Estrada, professor in the department of Computer Science at the University of New Mexico, Nikos Konstantinidis, professor of Physics at the University College London, and Jibo Sanyal, group leader of Computational Urban Sciences at the Oak Ridge National Laboratory connected from their home institutions. Maria Girone moderated the discussion and set the theme: “Learning Versus Understanding: Is AI a Definitive Game Changer in Science?”. Or how we can harness AI for humankind? The panelists were addressing the complexity of AI as well as some important ethical aspects. They agreed that an ethical AI paradigm requires user trust which means understanding bias. Not only obvious bias of which we are fully aware, such as gender, age, and race, to just name a few. Also hidden bias that AI will magnify inadvertently and unavoidably. When wondering what problems AI will never be able to solve, François Fleuret nicely but critically pointed out that AI will never come up with E=mc2, clearly hinting at the fundamental difference between artificial and human intelligence, between trained machines and genius and creativity. Human discovery will continue for as long as humans will exist!
The beauty of an interdisciplinary conference
In the various minisymposia ranging from computer science, to weather and climate, to physics, to materials science, to geoscience, and more, several important points were brought to light that are in need of further thought and consideration: Among them is the importance of continuing investment in performance portability, reflecting the diverse range of modern-day architectures. With the demise of Moore’s law, it appears that there is no longer low hanging fruit in High Performance Computing (HPC) or no fruit at all. Could new scientific areas like Machine Learning actually come to help? We still need to ensure that relevant scientific codes are running efficiently and at high performance and tap the potential of accelerated systems. Weather and climate research has gone through an enormous effort of porting codes to GPU accelerators; but we now see that the investment is paying off.
The challenges that the HPC community are facing are beyond the traditional territory of scientific computing. Scientists are now concerned on how HPC and data science could help addressing natural disasters like forest fires, earthquake damages, tsunami floods, all of which are extremely challenging but also very relevant use cases. Complex simulations throw up new problems, however: How difficult is it to reproduce natural events in silico; how reproducible and statistically significant are simulations; how do we ensure that obtained results are really answering the questions being asked? Some of the presented work focused on how to make best use of powerful compute and data infrastructure, how to recognize bottlenecks and how to improve work- and dataflows.
It appears that the paths to knowledge are still far from perfection and that there is still a lot of uncertainty in the models to be used. Proper use of all available domain knowledge is critical if we want AI, Deep Learning, and Machine Learning give meaningful results.
Double the fun!
PASC20 was postponed after submission of papers and reviewing process had been completed, and so two best paper awards could be announced, one for last year and one for this year. The PASC20 best paper award went to Md Bulbul Sharif and his co-workers at Tennessee Tech University in the USA for their study of performance evaluation of a two-dimensional flood model on heterogeneous High-Performance Computing architectures. The PASC21 best paper award went to Rahul Bale and his colleagues at RIKEN in Japan for their simulations of droplet dispersion to understand how an infected person could spread the SARS-CoV-2 virus. A magnificent Digital Event Portal allowed participants to stroll around virtually in the huge hall of University of Geneva, be present at the Poster Sessions, check out posters, and interact with presenters in real time. The best poster was selected by electronic poll among all participants, and the award went to Sara Faghih-Naini at University of Bayreuth in Germany for her poster on “Quadrature-Free Discontinuous Galerkin Method for Shallow-Water Equations”. Faghih-Naini connected remotely and received congratulations from the co-chairs of the PASC21 conference.
What would life be without some magic!
The PASC21 participants and their families were treated to a nice surprise, which hopefully helped to compensate for the lack of social events and interactions due to the online format of the conference. On Thursday evening, after another long and exciting day of scientific discussions, Federico Soldati, a professional artist, mentalist, and magician, ensorcelled the audience with magic tricks and with the illusion that he could predict the future or foresee the participants’ thoughts.
PASC21 out, PASC22 in…
After PASC20 had to be postponed due to an unexpected and ravaging pandemic, the PASC organizing committee and their two chairs Sinéad Ryan and Bastien Chopard could finally reap the benefits of two years of work and enjoy the fruit of their labor. These two years have not been easy. Going from an in-presence to a virtual event and retaining the charm and zing of a “real” conference was certainly a challenge. The organizers proved that they had mastered the challenge and put together an exciting interdisciplinary conference with an amazing Digital event portal. Important questions were addressed by the scientific community, new scientific challenges were revealed, and new incentives were laid out for the diverse communities to work together.
The torch was finally passed to the co-chairs of PASC22: Michela Taufer, professor in High-Performance Computing at the University of Tennessee, USA, and Petros Koumoutsakos, professor of Computing in Science and Engineering at Harvard University, USA and ETH Zurich, Switzerland. Jointly they invited participants to save the dates for next year, June 27-29, 2022 in Basel, Switzerland. The audience was left with inspiring words by Petros Koumoutsakos, who introduced the PASC22 theme “Computing and Data … for all Humankind” and revealing that the theme is inspired by the documentary film “For All Mankind”, which celebrates the Apollo Program that brought the first man to the Moon. This indeed lets us expect an amazing and exciting PASC22.