The Invited Talks for SC20 represent the breadth, depth and future outlook of technology and its societal and scientific impact. HPC has always played a critical role in advancing breakthroughs in weather and climate research. This year’s invited talks extend this further to data driven approaches, including biodiversity, geoscience, and quantum computing. Our speakers will also touch on responsible application of HPC and new technological developments to highlight the impact of this potent and versatile technology on a wide range of applications.
Hear these illustrious speakers during SC20 Invited Talks, Tuesday–Thursday, November 17–19.
Lorena Barba (George Washington University) will explore the need for trustworthy computational evidence through transparency and reproducibility. With the explosion of new computational models for vital research, including COVID-19, applications that are of such importance to society highlight the requirement of building trustworthy computational models. Emphasizing transparency and reproducibility have helped us build more trust in computational findings. How should we adapt our practices for reproducibility to achieve “unimpeachable provenance”, and reach full accountability of scientific evidence produced via computation?
Shekhar Borkar (Qualcomm Inc.) will speak on the future of computing in the so-called “post Moore’s law era.” While speculations about the end of Moore’s law have created some level of fear in the community, this ending may not be coming as soon as we think. This talk will revisit the historic predictions of ‘the end’, and discuss promising opportunities and innovations that may further Moore’s law and continue to deliver unprecedented performance for years to come.
Dalia A. Conde (University of Southern Denmark) will offer a presentation on fighting the extinction crisis with data. With biodiversity loss identified by the World Economic Forum as one of humanity’s greatest challenges, computational methods are urgently needed to secure a healthier planet. We must design and implement effective species conservation strategies, which rely on vast and disparate volumes of data, from genetics and habitat to legislation and human interaction. This talk will introduce the Species Knowledge Index initiative, which aims to map, quantify, analyze, and disseminate open information on animal species to policy makers and conservationists around the globe.
Tom Conte (Georgia Tech) will examine HPC after Moore’s law. Whether Moore’s law has ended, is about to end, or will never end, the slowing of the semiconductor innovation curve has left the industry looking for alternatives. Different approaches, beyond quantum or neuromorphic computing, may disrupt current algorithms and software development. This talk will preview the road ahead, and suggest some exciting new technologies on the horizon.
Marissa Giustina (Google LLC) will share the challenges and recent discoveries in the development of Google’s Quantum computer, from both the hardware and quantum-information perspectives. This prototype hardware holds promise as a platform for tackling problems that have been impossible to address with existing HPC systems. The talk will include recent technological developments, as well as some perspective for the future of quantum computing.
Patrick Heimbach (The University of Texas at Austin) will discuss the need for advanced computing to help solve the global ocean state estimation problem. Because of the challenge of observing the full-depth global ocean circulation in its spatial detail, numerical simulations play an essential role in quantifying patterns of climate variability and change. New methods that are being developed at the interface of predictive data science remain underutilized in ocean climate modeling. These methods face considerable practical hurdles in the context of HPC, but will be indispensable for advancing simulation-based contributions to real world problems.
Simon Knowles (Graphcore) will discuss the reinvention of accelerated computing for artificial intelligence. As HPC changes in response to the needs of the growing user community, AI can harness enormous quantities of processing power – even as we move towards power-limited computing. To balance these needs, the intelligence processor (IPU) architecture is able to capture learning processes and offer massive heterogeneous parallelism. This ground-up reinvention of accelerated computing will show considerable results for real applications.
Ronald P. Luijten (Data Motion Architecture and Consulting GmbH) will offer a presentation on data-centric architecture of a weather and climate accelerator. Using a co-design approach, a non-Von-Neumann accelerator targeting weather and climate situations was developed in tandem with the application code to optimize memory bandwidth. This also led to the filing of a patent for a novel CGRA (Course Grain Reconfigurable Array) layout that reflects grid points in the physical world. The talk will include benchmarks achieved in the project, and a discussion of next steps.
Catherine (Katie) Schuman (Oak Ridge National Laboratory) will introduce us to the future of AI and HPC, in the form of neuromorphic computing and neural accelerators. These two new types of computing technologies offer significant advantages over traditional approaches, including considerably increased energy efficiency and accelerated neural network-style computing. This talk will illustrate the fundamental computing concepts involved in these new hardware developments, and highlight some initial performance results.
Compton Tucker (NASA Goddard Space Flight Center) will speak on satellite tree enumeration outside of forests at the Fifty Centimeter Scale. Non-forest trees, which grow isolated outside of forests, and are not well documented, nevertheless play a crucial role for biodiversity, carbon storage, food resources, and shelter for humans & animals. This talk will detail the use of HPC and machine learning to enumerate isolated trees globally, to identify localized areas of degradation, and quantify the role of isolated trees in the global carbon cycle.
Cliff Young (Google LLC) will entertain the question of whether we can build a virtuous cycle between machine learning and HPC. While machine learning draws on many HPC components, the two areas are diverging in precision and programming models. However, it may be possible to construct a positive feedback loop between them. The Tensor Processing Unit (TPU) could provide opportunities to unite these fields to solve common problems through parallelization, mixed precision, and new algorithms.
Melyssa Fratkin, SC20 Communications Chair