The Center for Artificial Intelligence Innovation at the National Center for Supercomputing Applications at the University of Illinois won two of out the four NCSA-related awards at SC21. CAII won Best Use of High Performance Data Analytics & Artificial Intelligence and Best Use of HPC Physical Sciences. Both of these awards involved the use of the HAL system, which is run by CAII.
Editor’s Choice: Best Use of High Performance Data Analytics & Artificial Intelligence. A team led by Argonne National Laboratory, with scientists from NCSA and the University of Chicago, used the Summit supercomputer at Oak Ridge National Laboratory, NCSA’s HAL cluster, funcX, Nvidia’s TensorRT and the Data and Learning Hub for Science (DLHub) to create reproducible, physics-inspired AI models that search for and detect gravitational waves in one-month’s worth advanced LIGO data within seven minutes. It’s the first production-scale AI method that combines HPC and edge computing and enables orders of magnitude faster AI-driven inference of gravitational waves (published in Nature Astronomy).
Reader’s Choice: Best Use of HPC in Physical Sciences. An international scientific collaboration developed the first class of AI models that outperform humans in the classification of thousands of compact star clusters observed by the Hubble Space Telescope. The collaboration used NCSA’s HAL cluster in addition to systems at other institutions. Participating institutions are: Space Telescope Science Institute, Gemini Observatory, Caltech/IPAC, University of Chicago, Argonne National Laboratory, University of Illinois, Argelander-Institut für Astronomi, Zentrum für Astronomie der Universität Heidelberg, Australian National University, University of Alberta and Max Planck Institut für Astronomie. Resources used: AWS, Bridges-2 (Pittsburgh Supercomputing Center, allocated and supported by XSEDE), HAL (NCSA), NVIDIA cuDNN-accelerated deep-learning framework, TensorRT, and the Data and Learning Hub for Science.
Read more on NCSA’s HPC Wire awards here.
HPC Wire press release.