Skip to content

Members of the “CoE RAISE” EU undertaking – creating AI approaches for next-generation supercomputers – meet at CERN

Final week, members of the EU’s CoE RAISE undertaking met at CERN for his or her “All Fingers” assembly. This progressive undertaking is creating artificial-intelligence (AI) approaches for next-generation “exascale” supercomputers, to be used throughout each science and trade. Use instances explored by the undertaking embrace the optimization of wind-farm layouts, design of environment friendly plane, improved sound engineering, seismic imaging with distant sensing, and extra.

CoE RAISE – the European Heart of Excellence in Exascale Computing “Analysis on AI- and Simulation-Primarily based Engineering at Exascale” – is funded below the EU’s Horizon 2020 analysis and innovation programme. The undertaking launched in 2021 and runs for 3 years.

The four-day assembly, which came about in CERN’s Council Chamber, was attended by 54 undertaking members. The members mentioned progress made of their work to develop AI applied sciences for complicated functions in Europe operating on future “exascale” high-performance computing (HPC) programs. Exascale refers back to the subsequent era of high-performance computer systems that may perform over 1018 floating-point operations per second (FLOPS). As we speak, solely the Frontier supercomputer at Oak Ridge Nationwide Laboratory in the US has reached this stage. Nonetheless, with extra exascale HPC programs simply over the horizon, it is very important make sure that AI approaches utilized in science and trade are able to capitalize totally on the large potential. In June, the European Excessive Efficiency Computing Joint Endeavor (EuroHPC JU) introduced that Forschungszentrum Jülich GmbH in Germany has been chosen to host and function Europe’s first exascale supercomputer, which is ready to return on-line subsequent yr and will probably be often known as JUPITER (the Joint Endeavor Pioneer for Progressive and Transformative Exascale Analysis).

CoE RAISE is creating progressive AI strategies on heterogeneous HPC architectures involving a number of sorts of processor. Such architectures can provide larger efficiency and power effectivity, however code should be tailored to make use of the various kinds of processors effectively. The AI ​​strategies being developed are targeted round 9 key use instances and designed to scale properly for operating on exascale HPC programs.

CoE RAISE helps know-how switch to trade, significantly small- and medium-sized enterprises, in addition to operating schooling and coaching initiatives. On prime of this, CoE RAISE additionally gives consulting and liaises with different European initiatives to maximise synergies, exploit alternatives for co-design and share information. All facets of the undertaking’s work have been mentioned over the 4 days at CERN.

CERN can be a associate and brings one of many use instances to the undertaking. This work focuses on the advance of strategies for reconstructing particle-collision occasions on the upgraded Excessive-Luminosity Giant Hadron Collider (HL-LHC), which is ready to return on-line in 2029. The HL-LHC will see extra particle collisions than ever going down, producing exabytes of information annually, leading to unprecedented computing challenges. To reconstruct particle collision occasions at the moment (with information units within the order of terabytes or petabytes), a whole lot of various algorithms run concurrently: some are conventional algorithms optimized for specific {hardware} configurations, whereas others already embrace AI-driven strategies, akin to deep neural networks (DNNs). The members of the undertaking crew at CERN are working to extend the modularity of programs and make sure that code is optimized to totally exploit heterogeneous architectures, in addition to rising using machine studying and different AI strategies for reconstruction of collisions and classification of particles.

“Supercomputers are reaching the exascale and enabling the supply of an unprecedented scale of processing sources for HPC and AI workflows,” says Maria Girone, CERN openlab CTO, who leads CERN’s contribution to the undertaking. “The analysis carried out in CoE RAISE will drive the co-design of HPC computing sources for future AI and HPC functions for each science and trade. This assembly enabled us to change and develop concepts and to carry new views. It additionally gave researchers from different domains a singular perception into the surroundings and challenges going through CERN, selling cross-fertilization and understanding.”


Leave a Reply

Your email address will not be published. Required fields are marked *