Category II: Exploring Neural Network Processors for AI in Science and Engineering

点击翻译按钮获取中文摘要

Artificial Intelligence (AI) is increasingly applied to science and engineering problems, and as a part of a growing data science field that is enabling new insights and discoveries not possible with traditional high-performance computing architectures. In 2019 the President issued an executive order announcing a national AI strategy involving the private sector, academia, and the public. Among other things, this strategy calls for investments in AI research and development, providing AI resources, and training an AI-ready workforce. Researchers are increasingly applying machine learning (ML) techniques to science and engineering problems including those from astronomy, climate modeling, extreme-scale systems management, cell biology, high energy physics, drug discovery, social science, satellite image analysis, among others. In addition to advances in algorithms and software, the performance of AI systems is heavily dependent on the underlying hardware. Evaluation of hardware optimized for AI algorithms is of keen interest to the AI research community.To extend the application of AI to evermore challenging problems in science and engineering, the San Diego Supercomputer Center (SDSC), working closely with their vendor partner Supermicro, will deploy Voyager, a high-performance, innovative resource for conducting AI research across a wide range of science and engineering domains. Based on AI processors optimized for deep learning (DL) operations, Voyager will be a first-of-its-kind system available in the NSF resource portfolio. This will give researchers the opportunity to explore Voyager’s unique hardware and software using well-established deep learning frameworks like PyTorch, Keras, and Tensorflow to implement deep learning techniques such as convolutional neural networks (CNNs) and generative adversarial networks (GANs). Researchers will also be able to develop their own AI techniques using software tools and libraries built specifically for Voyager’s innovative AI architecture.The project is structured as a three-year Testbed phase followed by a two-year Allocations Phase. During the Testbed phase SDSC researchers and collaborators will work closely with a small number of research teams to evaluate the performance of Voyager’s innovative deep learning (DL) hardware, specialized compilers, and system libraries. Semiannual workshops will bring teams together to share lessons learned, and develop the knowledge and best practices that inform future users who will be given access during the Allocations Phase. During the Allocations Phase, Voyager will be available to researchers with projects deemed meritorious by an NSF-approved allocation process. Lessons learned from the Testbed Phase is used to develop documentation, best practices, allocations models, and user support strategies. Semiannual workshops continue in the Allocations Phase. Through SDSC’s AI Technology Lab, the project will engage with industry to explore how technologies like those in Voyager can improve the global competitiveness of private sector companies, and prepare the next generation workforce.SDSC will deploy Voyager in SDSC's energy-efficient data center on the UCSD campus. Voyager will be connected to multiple high-performance research and education networks at 100 Gbps. Supporting Voyager is a nationally recognized team of application and systems experts at SDSC. The Voyager External Advisory Board will assist in recruiting early users and providing guidance to the project.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.