Research on AI- and Simulation-Based Engineering at Exascale

Informacje na temat projektu

RAISE

Identyfikator umowy o grant: 951733

DOI

10.3030/951733

Projekt został zamknięty

Data podpisania przez KE 10 Grudnia 2020

Data rozpoczęcia 1 Stycznia 2021

Data zakończenia 30 Czerwca 2024

Finansowanie w ramach

• EXCELLENT SCIENCE - Research Infrastructures

Koszt całkowity

€ 4 969 347,04

Wkład UE

€ 4 969 347,04

Koordynowany przez

FORSCHUNGSZENTRUM JULICH GMBH
Germany

Periodic Reporting for period 2 - RAISE (Research on AI- and Simulation-Based Engineering at Exascale)

Okres sprawozdawczy: 2022-07-01 do 2024-06-30

Podsumowanie kontekstu i ogólnych celów projektu

The development of artificial intelligence (AI) methods is rapidly proceeding. Analyzing and processing big data require high computational power and scalable AI solutions. Therefore, it becomes mandatory to develop entirely new workflows from current applications that efficiently run on future high-performance computing (HPC) architectures at Exascale.

The European Center of Excellence in Exascale Computing "Research on AI- and Simulation-Based Engineering at Exascale" (CoE RAISE) started on Jan. 01, 2021, with a total budget of around €5 million. The project brings together eleven full partners and two third parties with expertise in AI and HPC. RAISE will be the excellent enabler for novel intertwined AI/HPC methods, which will be advanced along representative use cases, covering a wide spectrum of academic and industrial applications of societal importance, e.g. wind energy, wetting hydrodynamics, manufacturing, physics, turbomachinery, or aerospace.

CoE RAISE aims at closing the gap in full loops using forward simulation models and AI-based inverse inference models with novel hardware technologies, i.e. Modular Supercomputing Architectures and Quantum Annealing. Best practices, support, and education for industry, SMEs, academia, and HPC centers on Tier-2 level and below will be developed. They will be provided to RAISE's European network to attract new user communities for which a business will offer services around these new technologies.
The main objectives of CoE RAISE are to

• develop innovative AI methods on heterogeneous HPC architectures capable of scaling towards Exascale;
• generalize developments for selected representative simulation codes and data-driven workflows for universal application;
• ensure knowledge and technology transfer to industry, SMEs, and academia, and with respect to AI and HPC, to less developed countries and computing centers;
• provide corresponding training and education, and support, qualifying code and workflows for execution on heterogeneous/modular systems towards Exascale;
• create a European network of contact points to provide infrastructural and knowledge access, consulting, and further services to user communities from industry, SMEs, and academia with less developed expertise in AI and HPC;
• develop a business plan to ensure long-term sustainability, financial independence, and continuous extension of services beyond the horizon of the project;
• connect to existing European projects to exploit synergies, avoid redundancy, exploit co-design opportunities, and exchange knowledge.

Prace wykonane od początku projektu do końca okresu sprawozdawczego oraz najważniejsze dotychczasowe rezultaty

The members of CoE RAISE's WP2 jointly developed AI technologies with the use case teams from WP3 and WP4. The advancements of AI methods to accelerate computations and data processing intend to model physics or processes that otherwise require large computing power. Newest HPC architectures available through various production and prototype machines were used. Simulation and AI tools have been ported to these machines and corresponding performance analyses and improvements have been performed. Open-access documentation has been provided. The software layout for the open-source Unique AI Framework (UAIF) of CoE RAISE has been designed.

The smaller HPC centers from RWTH and RTU were included in the development works of the UAIF components, some of which have already been made open-source. Furthermore, trainings (available on RAISE’s YouTube channel) have been organized on almost a monthly basis. Topics on AI and HPC were covered. The trainings included external contributions form industry, e.g. Graphcore, and from less HPC-oriented communities such as OpenML. The online Educational services platform and an educational portfolio have been developed. Advertisements happened over the website and social media channels in WP6. To establish a RAISE business, WP5 has performed a market analysis, which forms the base for a business plan.

Connections to the EuroCC project have been established to fertilize collaborations with national industry and SMEs. Connections to projects responsible for hardware, and the provided middle- and software stacks, were established, e.g. to the Dynamical Exascale Entry Platform (DEEP) project series. RAISE collaborates with CoE Combustion (CoEC) on a joint use case on hydrogen combustion and also with the Energy-oriented CoE (EoCoE-II) on developments of complementary methods for wind turbine simulations. CoE RAISE also became part of the HPC3 CoE Council, organized by FocusCoE, and established connections through this council.

Innowacyjność oraz oczekiwany potencjalny wpływ (w tym dotychczasowe znaczenie społeczno-gospodarcze i szersze implikacje społeczne projektu)

The AI and HPC experts of WP2 provide their expertise to the WP3 and WP4 domain experts, who require AI/HPC resources for data-intensive workflows. Generalizable AI tools such as hyperparameter tuning have found their way into the application workflows, thus accelerating training. The full loop implementations that would directly impact the model optimization workflow, are not yet available. It should, however, be noted that such an implantation is – on an experimental level – already available in RAISE. All partners were given the opportunity to test state-of-the-art HPC machines and prototypes through computing time provided by the HPC hosting entities of RAISE, or by the PRACE Rapid Access program.

Leveraging technical competitiveness for EU industries and SMEs through knowledge and technology transfer from research to the respective economic realms can be realized in the form of (i) guidance towards working solutions, e.g. by offering consulting services, (ii) co-development in a joint research project, (iii) solution-as-a-service offers, or (iv) trainings and dissemination of project results. The project has already worked on advancing points (ii) and (iv), e.g. CERFACS, BULL, and SAFRAN jointly work on hydrogen combustion and FM aims to impact Additive Manufacturing industries. Points (i) and (iii) will be covered by the RAISE business that is being developed in WP5.

Connecting existing European projects to exchange knowledge is one of the objectives of RAISE. Especially the common intention of the CoEs is to develop methods for the next generation of HPC systems. The application focus of the CoEs is, however, quite different. Despite the presence of AI in many CoEs, it is most of the time only used as a tool without exploiting HPC, and is not necessarily the main output of the CoE. This will change in the future with the advent of large data volumes. In contrast to other CoEs, RAISE considers itself as THE AI CoE that wants to provide scalable AI technologies and services. In this sense, the developments of RAISE already impact the developments in other European projects. Commonalities with TREX CoE are currently being investigated. FocusCoE further fertilized possible collaborations through HPC3 meetings. RAISE was active in workshops on Co-Design, in FocusCoE’s WP3 activities, and in joint EuroHPC Summit Week outreach activities. In the future, RAISE will further exploit its connections to other CoEs (especially new CoEs) to support them in applying scalable AI, and to advance codes to be Exascale-ready. For example, RWTH joins the upcoming EXCELLERAT2 project (it participated also in EXCELLERAT), this time working on the m-AIA code, which is a RAISE reference code. A collaboration of the two CoEs is hence only natural. To summarize, the developments of RAISE and other CoEs complement one another and allow all projects to mutually benefit.

Geographical distribution of CoE RAISE's partners and their interactions.

Impact hierarchy of CoE RAISE gathered by the stakeholders.

CoE RAISE's compute- and data-driven use cases.

CoE RAISE's contributions to AI at Exascale.