HDR Institute: Accelerated AI Algorithms for Data-Driven Discovery

HDR 研究所：用于数据驱动发现的加速 AI 算法

• 批准号: 2117997
• 负责人: Shih-Chieh Hsu
• 金额: $ 1500万
• 依托单位: University of Washington
• 依托单位国家: 美国
• 项目类别: Cooperative Agreement
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-10-01 至 2026-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2117997&HistoricalAwards=false
• 关键词: HDR; Institute; Accelerated; AI; Algorithms

The data revolution is dramatically accelerating the acquisition rate of new information, creating a vast amount of data. Artificial intelligence (AI) has emerged as a solution for rapid processing of complex datasets. New hardware such as graphics processing units (GPUs) and field-programmable gate arrays (FPGAs) allow AI algorithms to be greatly accelerated. To take full advantage of fast AI, the Institute of Accelerated AI Algorithms for Data-Driven Discovery (A3D3) targets fundamental problems in three fields of science: high energy physics, multi-messenger astrophysics, and systems neuroscience. A3D3 works closely within these domains to develop customized AI solutions to process large datasets in real-time, significantly enhancing their discovery potential. The ultimate goal of A3D3 is to construct the institutional knowledge essential for real-time applications of AI in any scientific field. Through dedicated outreach efforts, A3D3 will empower scientists with new tools to deal with the data deluge. Students mentored through A3D3 research will interact closely with industry partners, creating new career opportunities and strengthening synergies between academia and industry.The approach of A3D3 is to tightly couple AI algorithm innovations, heterogeneous computing platforms, and science-driven application development informed through close collaboration with domain scientists within physics, astronomy, and neuroscience. The common theme across domains is the development of AI strategies accelerated by emerging processor technology, employing hardware-AI co-design as a transformative solution to a wide range of scientific challenges. Hardware architectures such as GPUs and FPGAs have emerged as promising technologies to address many of the challenges in data-intensive science because they provide highly-performant, parallelizable, and configurable data processing pipeline capabilities. When combined with AI algorithms, these architectures significantly accelerate scientific workflows compared to CPU-only computing platforms. Building on the existing Fast Machine Learning community, A3D3 cultivates an ecosystem where scientists across domains collaborate to meet critical challenges, forming a central hub of excellence for innovation in accelerated AI for science. The work is extended to the public at large through a diverse set of educational training programs and by mentoring next-generation scientists.This project is part of the National Science Foundation's Big Idea activities in Harnessing the Data Revolution (HDR) and Windows on the Universe - The Era of Multi-Messenger Astrophysics (WoU-MMA). This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Divisions of Astronomical Sciences and of Physics within the NSF Directorate for Mathematical and Physical Sciences.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.

数据革命极大地加快了新信息的获取速度，产生了大量的数据。人工智能（AI）已经成为快速处理复杂数据集的解决方案。 图形处理单元（GPU）和现场可编程门阵列（FPGA）等新硬件使AI算法大大加速。 为了充分利用快速人工智能，数据驱动发现加速人工智能算法研究所（A3 D3）针对三个科学领域的基本问题：高能物理学，多信使天体物理学和系统神经科学。 A3 D3在这些领域密切合作，开发定制的人工智能解决方案，以实时处理大型数据集，显著提高其发现潜力。 A3 D3的最终目标是构建人工智能在任何科学领域的实时应用所必需的机构知识。通过专门的推广工作，A3 D3将为科学家提供新的工具来处理数据泛滥。 通过A3 D3研究指导的学生将与行业合作伙伴密切互动，创造新的职业机会并加强学术界和工业界之间的协同效应。A3 D3的方法是将人工智能算法创新、异构计算平台和科学驱动的应用程序开发紧密结合起来，通过与物理学、天文学和神经科学领域的领域科学家密切合作。 各领域的共同主题是新兴处理器技术加速了人工智能战略的发展，采用硬件-人工智能协同设计作为应对各种科学挑战的变革性解决方案。 GPU和FPGA等硬件架构已成为解决数据密集型科学中许多挑战的有前途的技术，因为它们提供了高性能，可并行和可配置的数据处理管道功能。 当与AI算法相结合时，与仅使用CPU的计算平台相比，这些架构显著加快了科学工作流程。 基于现有的快速机器学习社区，A3 D3培养了一个生态系统，跨领域的科学家合作应对关键挑战，形成了加速人工智能科学创新的卓越中心。 该项目是美国国家科学基金会在利用数据革命（HDR）和宇宙之窗--多信使天体物理学时代（WoU-MMA）中的大创意活动的一部分。 该奖项由高级网络基础设施办公室颁发，由NSF数学和物理科学理事会的天文科学和物理部门共同支持。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Interpretable Geometric Deep Learning via Learnable Randomness Injection

• DOI: 10.48550/arxiv.2210.16966
• 发表时间: 2022-10
• 期刊: ArXiv
• 作者: Siqi Miao;Yunan Luo;Miaoyuan Liu;Pan Li

Unveiling hidden physics at the LHC

• DOI: 10.1140/epjc/s10052-022-10541-4
• 发表时间: 2021-09
• 期刊: The European Physical Journal C
• 作者: O. Fischer;B. Mellado;S. Antusch;E. Bagnaschi;S. Banerjee;G. Beck;Benedetta Belfatto;M. Bellis;Z. Berezhiani;M. Blanke;B. Capdevila;K. Cheung;A. Crivellin;N. Desai;Bhupal Dev;R. Godbole;T. Han;P. Harris;M. Hoferichter;M. Kirk;S. Kulkarni;C. Lange;K. Lassila-Perini;Zhen Liu;F. Mahmoudi;C. A. Manzari;D. Marzocca;B. Mukhopādhyāẏa;A. Pich;Yifeng Ruan;Luc Schnell;J. Thaler;S. Westhoff

TorchSparse++: Efficient Point Cloud Engine

• DOI: 10.1109/cvprw59228.2023.00025
• 发表时间: 2023-06
• 期刊: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)
• 作者: Haotian Tang;Shang Yang;Zhijian Liu;Ke Hong;Zhongming Yu;Xiuyu Li;Guohao Dai;Yu Wang;Song Han

Algorithm and System Co-design for Efficient Subgraph-based Graph Representation Learning

• DOI: 10.14778/3551793.3551831
• 发表时间: 2022-02
• 期刊: Proc. VLDB Endow.
• 作者: Haoteng Yin;Muhan Zhang;Yanbang Wang;Jianguo Wang;Pan Li

ScaleHLS: a scalable high-level synthesis framework with multi-level transformations and optimizations: invited

ScaleHLS：具有多级转换和优化的可扩展高级综合框架：受邀

• DOI: 10.1145/3489517.3530631
• 发表时间: 2022
• 期刊: ACM
• 作者: Ye, Hanchen;Jun, HyeGang;Jeong, Hyunmin;Neuendorffer, Stephen;Chen, Deming
• 通讯作者: Chen, Deming