Le lien vers la présentation.

Bio

Dr. L. Gicquel détient un doctorat en mécanique des fluides et énergétique de l’Université de New York à Buffalo (2001). Il a ensuite rejoint le CERFACS pour être récemment nommé Project Leader de l’équipe « Energy & Safety » anciennement équipe « Combustion ».  Tout au long de ces 23 ans au CERFACS, il a pu contribuer au développement de la Simulation aux Grandes Echelles (SGE), que ce soit au travers de l’outil AVBP, la modélisation de la turbulence, des écoulements turbulents et réactifs ou plus largement l’utilisation de la SGE appliquée aux problèmes de l’industrie.

Le lien vers la présentation.

Le lien vers la présentation.

Résumé

La recherche en fusion est en pleine effervescence. Divers concepts ont été proposés pour confiner des plasmas extrêmement chauds (plusieurs centaines de millions de degrés) suffisamment longtemps pour produire de l’énergie à partir de réactions de fusion. Parmi ces concepts, le tokamak est actuellement le plus prometteur et a été choisi pour le projet international ITER en construction sur le site de Cadarache. Le succès de ce projet passe notamment par la compréhension et la prédiction du transport turbulent de particules et de chaleur au sein du plasma.

Les plasmas de tokamak sont peu collisionnel et nécessitent donc une description cinétique. De plus, le confinement du plasma augmente fortement avec le rapport entre la taille de la machine et la taille des structures turbulentes. Hors une bonne description du confinement nécessite de simuler l’ensemble des échelles présentes. Ainsi le coût numérique d’une simulation d’un plasma de tokamak est important et croît avec sa qualité.

Le code GYSELA, développé au CEA depuis plus de 20 ans, permet de modéliser le transport turbulent au sein des plasmas de tokamaks. Nous verrons à travers quelques exemples comment l’augmentation de la puissance de calculs des 10 dernières années a permis d’augmenter considérablement la précision des prédictions sur des tokamaks de taille actuel et pourquoi des machines Exascale seront nécessaire pour modéliser des plasmas de la taille d’ITER.

Bio

Peter Donnel est chercheur à l’Institut de Recherche sur la Fusion Magnétique (IRFM) du CEA depuis 2021. Après des études de physique à l’ENS Ulm, il a réalisé sa thèse (2015-2018) également à l’IRFM, suivi d’un post-doc à l’EPFL. Il travaille sur la description théorique et numérique des plasmas de tokamaks.

Le lien vers la présentation.

Abstract

In recent decades, a revolution has taken place in biology, thanks to major scientific advances brought about by the new generation of sequencing techniques. This « omics revolution » is accompanied by new challenges in computer science, in particular in managing huge amounts of data, which benefit from the most efficient approaches in artificial intelligence. For example, it is now possible to better understand the origin of pathologies by identifying mutations in genes, or to follow the evolution of viral strains associated with different virulence characteristics. A detailed molecular view and interpretation is now possible, allowing chemists and biologists to work together and provide valuable information on crucial biological questions.

The COVID-19 pandemic is an emblematic example of such a partnership. Theoretical chemists, in particular, have provided important insights into the mechanisms of interaction between the viral spike protein and the host receptor. One of the approaches they used was molecular dynamics simulation, which allowed them to go beyond the classical quasi-static view provided by X-ray or electron microscopy. Thanks to the detailed view provided by molecular dynamics simulations, the design of new drugs can be carried out in a more rational way. This approach in itself can greatly speed up drug development.

This is only the most famous example, but applications to major health problems such as antibiotic resistance, neurodegenerative diseases or cancer, to name but a few, have also been important success stories. I will present some of these success stories, written thanks to these theoretical methods.

I will also highlight how such approaches require high-performance computer facilities that should be constantly upgraded to be competitive and bring valuable results.

Le lien vers la présentation.

Abstract

In recent years, machine learning has revolutionized materials science, driven by the maturity of numerous algorithms and the exponential growth of computing power. These advances have had a profound impact on computational materials research and are increasingly influencing experimental characterization techniques. As the head of the CNRS research network « AI and Materials (IAMAT), » I will present recent developments in the application of machine learning to materials science, covering both theoretical/simulation aspects and experimental applications. The presentation will showcase how AI-based approaches are being used to explore large datasets, enhance atomistic simulations, and improve the characterization of materials through techniques such as spectroscopy and imaging, highlighting the transformative potential of AI in this field.

Bio

Marco Saitta is a Professor of Exceptional Class 2 in Physics at Sorbonne Université. He obtained a PhD in Condensed Matter Theory from the International School for Advanced Studies in Trieste in 1997, and had then a postdoctoral position at the University of Pennsylvania of Philadelphia. In 2000, he was appointed Maître de Conférences at Université Pierre et Marie Curie, now merged into Sorbonne Université, at the Institut de Minéralogie, de Physique des Matériaux et de Cosmochimie (IMPMC), where he acceded to professorship in 2011.

He is since 2021 the director of IMPMC (about 120 permanent staff), director of the French national network Groupement de Recherche (GdR) “Intelligence Artificielle en sciences des MATériaux (IAMAT)” (more than 700 researchers), and PI of the Sorbonne University consortium “MAESTRO (Materials for Energy)”. A specialist of electronic structure theory and ab initio calculations, his research activity has spanned from bulk semiconductors to graphene and nanotubes, to water and ices, his main interests being the exotic properties of molecular crystals, liquids and amorphous at extreme conditions of pressure and temperature.

In recent years his research has opened up into more interdisciplinary fields, such as Earth sciences and biogeochemistry, with a special focus since 2020 on the development and use of machine learning methods in materials science. He has authored more than 120 articles, cited more than 5000 citations, and with a h-index = 44 (WoS). He has received more than 40 invitations to conferences.

Le lien vers la présentation.

Résumé

Le calcul intensif est un allié de premier plan pour aider à la compréhension de phénomènes physiques complexes qui régissent le monde qui nous entoure. Il est en constante évolution et a connu ces dernières années des progrès spectaculaires, aidé par l’arrivée des super calculateurs. Ces progrès sont réalisés grâce à la combinaison d’algorithmes mathématiques de pointe qui permettent d’utiliser au mieux les architectures de calcul moderne. Cet  exposé sera l’occasion de décrire quelques résultats récents remarquables qui ont bénéficié des moyens de calcul mis à disposition par les centres de calcul français, qu’ils soient publics ou privés.

Le lien vers la présentation.

Le lien vers la présentation.

En ouverture de cette journée, le comité scientifique d’ORAP propose un petit retour en arrière, comparant les perspectives du forum des 20 ans d’ORAP aux dix dernières années dans le domaine du calcul haute performance.

Le lien vers la présentation.

Résumé

Dans cette présentation, j’exposerai l’histoire de PRACE et ses principaux succès. Pendant plus d’une dizaine d’année, PRACE a marqué l’écosystème européen du HPC. Durant cette période, de nombreux projets scientifiques et industriels ont été soutenus et des services ont été déployés avec un haut niveau d’exigence, par exemple en termes de support aux utilisateurs et de formation. PRACE s’est forgé une place de premier plan dans la communauté HPC mondiale et continuera à apporter une contribution importante en tant qu’association d’utilisateurs et de centres HPC en Europe.

Bio

Florian Berberich est chef du bureau de l’UE et de la coopération internationale au Jülich Supercomputing Centre (Forschungszentrum Jülich). Il est également directeur des opérations de PRACE aisbl depuis 2020. En octobre 2015, il est devenu membre du conseil d’administration de PRACE aisbl. Il a travaillé pour le bureau de gestion du projet PRACE au Jülich Supercomputing Centre, depuis 2008. Il a obtenu son doctorat en physique à l’université technique de Dresde en 2002. Il a travaillé comme post- doc à l’ESRF en France avant de devenir assistant du conseil d’administration du Forschungszentrum Jülich en 2004.

Le lien vers la présentation.

Résumé

Créée par les pouvoirs publics en 2007, GENCI (Grand Équipement National de Calcul Intensif) est une grande infrastructure de recherche, société civile et opérateur public visant à démocratiser l’usage de la simulation numérique par le calcul haute performance associé à l’usage de l’intelligence artificielle et du calcul quantique pour soutenir la compétitivité́ scientifique et industrielle française. Il est en charge de l’implémentation des stratégies nationales en HPC, en IA (AIforHumanity) et dans les technologies quantiques pour le calcul (cadre de la SNQ). Genci contribue  enfin au renforcement des capacités européennes de calcul dans le cadre de l’initiative EuroHPC .

Philippe Lavocat présentera l’actualité des projets au niveau national, régional et européen notamment répondant à la stratégie HPC/IA de la machine de classe Exascale Alice Recoque, les projets quantiques HPCQS et EuroQCS de l’initiative nationale HQI et les perspectives  à moyen terme.

Bio

Après avoir passé plus de 30 ans à exercer des responsabilités dans des grands programmes internationaux (CERN, Fermilab, ESO, CNES, ESA, NASA, Iter) au CEA et au CNRS puis au Ministère de la Recherche (Chef du Service de la stratégie de la recherche ) autour des très grandes infrastructures de recherche (ILL, ESRF, ESS, XFEL, EMBL), Philippe Lavocat a été nommé PDG de GENCI (Grand Equipement National de Calcul Intensif) le 1er juillet 2016. En juin 2020, il devient Vice-Chairman du PRACE Council (PRACE- PartneRship for Advanced Computing in Europe), infrastructure de recherche européenne de l’ESFRI (European Scientific Forum for Research Infrastructure). Il est membre observateur du Conseil Administration du CERFACS depuis 2018 et du Conseil d’administration du CINES depuis 2020. Il est également depuis 2019 « Technical Advisor » auprès de la DGRI et de la DGE qui représentent la France dans le « Governing Board » de l’initiative européenne EuroHPC.

Le lien vers la présentation.

Abstract

The challenges of Exascale and the technological and scientific breakthrough it represents have been widely discussed in Europe and internationally for over 15 years. Some of the participants at the 30^th ORAP Forum may remember their involvement in the IESP and  BDEC projects, as well as in the European EESI and EXDCI contributions. Exascale systems are currently being installed in the USA and China, others will soon arrive in Europe, and post-exascale systems are already announced for Japan at the end of this decade. This presentation will detail some Exascale issues that remain (partially) unresolved, then discuss some challenges of the Post-Exascale age. We will focus in particular on four of these, which have been discussed as part of the French NumPEx program and of the international InPEx project:  “AI4HPC/HPC4AI”, “From the edge to HPC systems, the digital continuum », « Software, the new frontier » and « Co-design of software/applications ».

Bio

Jean-Yves Berthou has been appointed Executive Director at Inria, Delegate for Site Policy (Directeur Général Délégué Politique de Site, DGD PS) in October 2024. The DGD PS aims to strengthen the collective dynamics of higher education, research and innovation that form the basis of the Institute’s territorial strategy. As such, DGD PS coordinates and steer the implementation of the Institute’s site policy in all its dimensions: research, including European and international strategy, transfer and innovation, participation in training initiatives, articulation of support functions, and relations with local players. DGD PS’s mission is to embody Inria’s presence in the French territory through the coordination and the development of the Inria research centers in and through digital science and technology.

From February 2019 to September 2024, Jean-Yves Berthou was the director of the Inria Saclay research center. During this period, he set up the two components of the Inria Saclay research center, the Paris Saclay University Inria center and the Institut Polytechnique de Paris Inria center. During this period, the center has experienced strong growth. From 28 teams in 2019, the center now has 42 Inria research teams.

From 2015 to 2019, Jean-Yves was Director of Innovation for the EDF Business Unit Transformation and Operational Efficiency (6000 employees) which produces and operates IT, real estate, purchasing, and tertiary (HR, accounting, …) services and products for the entire EDF group. Jean-Yves set up a dynamic innovation approach, which has resulted in the industrialization of more than 250 innovations and nearly 70 partnership contracts with start-ups.

From 2011 to 2015, Jean-Yves was head of the Numerical and Mathematical Department (NuMa) at the French National Research Agency (ANR), in charge of organizing the selection, funding and monitoring of research projects as well as setting up European and international collaborations.

From 1997 to 2011, he held various positions within the research department of EDF, EDF R&D. He was successively researcher, project leader and manager before becoming head of the Simulation program and the Information Technology program of EDF R&D.

Jean-Yves Berthou holds a PhD in computer science from Sorbonne University. He is a specialist in simulation and high-performance scientific computing. He is co-director of the French Computing@(post)-Exascale program (NumPEx). For the past twenty years, Jean-Yves Berthou has been regularly called upon by the European Commission for expert missions, particularly in the fields of high-performance computing, open science and research and innovation strategies.

Le lien vers la présentation.

Le lien vers la présentation.

Abstract

We present some initial lessons, as well as technical and capacity challenges, from the Exa-DI co-design and co-development process across NumPeX of software components (libraries, frameworks, workflows, tools) driven by computational and communication motifs cross-cutting CSE application demonstrators and their integration into Motif-based proxy apps and Software Development Kits for advancing exascale scientific productivity through better scientific software development methodologies and sustainability.

Le lien vers la présentation.

Abstract

At a time when every public position is either the subject of instant « meme-ization » or sifted through by a fact-checking process that is as rigorous as it is often invisible, the emergence of high-performance text-generation tools of confounding ease of use raises questions, raises concerns and sometimes even worries. Most of these tools available to the general public are pure black boxes about which we know little or nothing: their training data? nothing. Their architectures? So little. Their performance? usually a matter of guesswork outside lab’s benchmarks.

Can we at least detect their content and therefore their possible influences? Not really, even OpenAI, creator of ChatGPT, reports success rates of 26%. In this presentation, I’ll give an overview of the main language models, question their relevance in an academic environment and address the question of their detectability in adversarial contexts. I’ll also address the growing problem of their weaponization.

Bio

Djamé Seddah is a tenured associate professor at Sorbonne University and on leave at INRIA Paris in the Almanach team. His interests cover the field of natural language processing, mainly wide-coverage multilingual syntactic analysis, the syntax-semantics interface, language models for low-resource languages, etc. A specialist in the construction of annotated corpora (Sequoia corpus, French Social Media Bank, French Question Bank, Narabizi Treebank, etc.), he participated in the development of the CamemBERT, PagnolXL and CamemBERTa language models, as well as character-based models for dialectal and highly noisy languages. His current research focuses on language models and possible ways of avoiding their weaponization (content detection, bias detection and mitigation, etc.).

Le lien vers la présentation.

Abstract

The SKA Observatory (SKAO) is in charge of constructing and operating the world’s largest low- to mid-frequency radio telescopes: SKA-Mid in South Africa and SKA-Low in Australia. Radio-telescopes fundamentally rely on massive data processing to form observables after the radio waves of cosmic origin have been digitized. The numbers of antennas and of frequency channels for SKA-Mid and SKA-Low lead to considerable amounts of raw data which first need to be combined to form beams or visibilities so that data reduction can be carried out to yield data products that astronomers can analyse. The design for the SKA has one processing facility for each of these stages. The first two, the Central Signal Processors (CSP) and the Science Data Processors (SDP) are part of the SKA-Mid and SKA-Low instruments themselves, while the SKA Regional Centres are organised as a federation of worldwide resources (SRCNet) forming the portal where the science data products are archived, post-procesed and analysed. In this talk, I will describe how the imperative of data flow reduction and the diversity of tasks shape this infrastructure. I will then focus on the SDP, which provides advanced signal processing and is driven by efficacy due to stringent financial and energetic constraints, and the emergent SRCNet with the opportunities it offers of continued or more advanced processing, albeit with lower yields due to a more complex environment. I will also discuss sustainability, « a foundational value of SKAO underpinning all other activities across the Observatory », which needs not be forgotten as the project strives for feasibility, and SKA France’s involvement in support it.

Bio

Shan Mignot is a CNRS research engineer in the Lagrange laboratory at Observatoire de la Côte d’Azur. After an early-career contribution to the software and hardware architecture for the data processing on board the Gaia satellite, he has contributed to the systems engineering and project management of GYES, 4MOST and MSE, all ground-based multi-object spectrographs. He has joined the SKA project 3 years ago and is the Product Owner of the SCOOP team in SKAO’s SAFe agile framework to drive the hardware-software co-design for the Science Data Processors. He also contributes to involving the academic community on the upstream R&D and challenges it in finding ethical ways to procure, operate and maintain sub-exascale machines on the timescale of SKA’s 50-year lifetime in such contexts as the NumPEx PEPR, the ECLAT joint laboratory and the preparation of the French contribution to the SRCNet.

Le lien vers la présentation.

Les jumeaux numériques consistent en une vaste panoplie de modèles et d’implémentations. D’après Gartner, ils ont atteint le sommet de leur “hype cycle” en 2018 et sont désormais reconnus comme étant trop complexes, trop couteux et non-transférables. Sommes-nous en fin de leur cycle de vie utile ? Dans ce talk je proposerai un cadre mathématique pour leur compréhension et leur implémentation qui peut traiter adéquatement la question de la complexité et faire le lien entre la modélisation classique et les approches “data-driven ». Ensuite, je discuterai de leur remplacement potentiel par des modèles d’IA générative.

Le lien vers la présentation « Reproductibilité ».

Le lien vers la présentation « Impact environnemental ».

La reproductibilité est au coeur de l’activité scientifique. Le calcul haute performance présente des spécificités qui interrogent sur la notion de reproductibilité. Les exposés du Forum Orap consacré à ce sujet ont permis de faire le point sur ces questions, en identifiant les problématiques et en s’appuyant sur des retours d’expérience de différentes communautés scientifiques.

Le lien vers la présentation.

Abstract

The first generation of exascale computing systems are online along with powerful new application capabilities and system software. At the same time, demands for high performance computing continue to grow for more powerful simulations, adoption of machine learning methods, and huge data analysis problems arising for new instruments and increasingly ubiquitous data collection devices.  In its broadest sense, computational science research is expanding beyond physical and life sciences into social sciences, public policy, and even the humanities.

With chip technology facing scaling limits and diminishing benefits of weak scaling, it will be increasingly difficult to meet these new demands.  Disruptions in the computing marketplace, which include supply chain limitations, a shrinking set of system integrators, and the growing influence of cloud providers are changing underlying assumptions about how to acquire and deploy future supercomputers.  At the same time there are discussions around cloud computing, specialized hardware, and the role of AI, both as a technique in science and as a driver of computing hardware and software.

In this talk I’ll present some of the findings of a US National Academies consensus report on the future of post-exascale computing, which states that business as usual will not be sufficient.  I will also give my own perspectives on some of the challenges and opportunities faced by the research community.

Bio

Katherine Yelick is the Vice Chancellor for Research at the University of California, Berkeley, where she also holds the Robert S. Pepper Distinguished Professor of Electrical Engineering and Computer Sciences. She is also a Senior Faculty Scientist at Lawrence Berkeley National Laboratory. She has been recognized for her research and leadership in high performance computing and is a member of the National Academy of Engineering and the American Academy of Arts and Sciences.

Le lien vers la présentation.