Collaborative Research: CCRI: New: A Scalable Hardware and Software Environment Enabling Secure Multi-party Learning

协作研究：CCRI：新：可扩展的硬件和软件环境支持安全的多方学习

• 批准号: 2213701
• 负责人: Heng Huang
• 金额: $ 48万
• 依托单位: University of Pittsburgh
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-10-01 至 2024-01-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2213701&HistoricalAwards=false
• 关键词: Collaborative; Research; CCRI; New; Scalable

Advances in machine learning have made a major impact on many real-world applications over the past decade, and have achieved scientific and engineering breakthroughs across many disciplines. A new era of collaborative learning is emerging as part of the next phase of ubiquitous computing, wherein researchers at different sites will work together to correlate the disparate data they have separately acquired and eventually create a sophisticated decision-making model. It is thus imperative to establish a platform to support collaborative, multi-party data analysis, through which the participating parties can share their data with each other with different degrees of privacy control. The participants can compute with each other's data, by either directly sharing data with the server or only sharing their model parameters with the server to collaboratively derive a solution with other parties. To make such an environment available to the community, this project establishes a scalable and trusted hardware and software environment, termed Bridge, to support a general form of collaborative machine learning. The Bridge platform enables scalable multi-party learning and data analysis in a variety of forms, in both centralized and decentralized settings, with security and privacy guarantees. The project's novelties are to synergistically design and integrate both hardware and software innovation as well as a suite of security and privacy mechanisms and tools to support various types of multi-party machine learning. The project's impacts are to enable collaborative research efforts in diverse communities of CISE researchers pursuing focused research agendas in computer and information science and engineering, and generate enormous social and economic benefits to individuals and organizations. The minority students and under-served populations will be engaged in research activities to create an inclusive environment where everyone contributes to and benefits from cutting-edge scientific research.The Bridge platform will develop a unified hardware and software infrastructure to achieve hardware and software co-design for multi-party learning. An algorithmic software infrastructure is designed to support distributed, federated, and multi-modal model learning and sharing. The Bridge platform integrates cryptographic (secure multi-party computation) and noise-based methods (differential privacy) to provide privacy across the entire process from data collection to output. The Bridge platform provides a set of tools on integrated data access, AutoML, team creation, machine learning model vulnerability evaluation, and heterogeneous feature embeddings to support flexible user applications. The Bridge platform ensures the scalability in the number of tasks, the number of users, and heterogeneity of data types by developing advanced techniques to improve asynchronous model updates, communication efficiency, fast convergence, and vertical data partition. The Bridge platform builds a collaborative learning community and accelerates many new research areas in the core Computer and Information Science and Engineering (CISE), such as advanced machine learning and data science, data privacy and trustworthy AI, convergent research among hardware, software and machine learning, and intelligent internet of things.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.

在过去十年中，机器学习的进步对许多现实世界的应用产生了重大影响，并在许多学科中实现了科学和工程突破。协作学习的新时代正在出现，作为无处不在计算的下一阶段的一部分，其中不同地点的研究人员将共同努力，将他们分别获得的不同数据关联起来，并最终创建一个复杂的决策模型。因此，必须建立一个平台来支持协作的多方数据分析，通过该平台，参与方可以在不同程度的隐私控制下相互共享数据。参与者可以通过直接与服务器共享数据或仅与服务器共享其模型参数来计算彼此的数据，以与其他方协作地导出解决方案。为了使这样的环境可供社区使用，该项目建立了一个可扩展和可信的硬件和软件环境，称为Bridge，以支持协作机器学习的一般形式。Bridge平台支持以各种形式进行可扩展的多方学习和数据分析，包括集中式和分散式设置，并具有安全性和隐私保证。该项目的创新之处在于协同设计和集成硬件和软件创新以及一套安全和隐私机制和工具，以支持各种类型的多方机器学习。该项目的影响是使CISE研究人员的不同社区的合作研究工作，追求重点研究议程在计算机和信息科学与工程，并产生巨大的社会和经济效益的个人和组织。通过搭建统一的软硬件基础设施，实现软硬件协同设计，实现多方学习。算法软件基础设施旨在支持分布式、联合和多模态模型学习和共享。Bridge平台集成了加密（安全多方计算）和基于噪声的方法（差分隐私），以在从数据收集到输出的整个过程中提供隐私。Bridge平台提供了一套关于集成数据访问、AutoML、团队创建、机器学习模型漏洞评估和异构特征嵌入的工具，以支持灵活的用户应用程序。Bridge平台通过开发先进的技术来提高异步模型更新、通信效率、快速收敛和垂直数据分区，确保了任务数量、用户数量和数据类型异构性的可扩展性。Bridge平台建立了一个协作学习社区，并加速了计算机与信息科学与工程（CISE）核心领域的许多新研究领域，如先进的机器学习和数据科学，数据隐私和可信的人工智能，硬件，软件和机器学习之间的融合研究，该奖项反映了NSF的法定使命，并通过使用基金会的智力价值进行评估，被认为值得支持和更广泛的影响审查标准。

项目成果

Challenges and Opportunities to Enable Large-Scale Computing via Heterogeneous Chiplets

• DOI: 10.1109/asp-dac58780.2024.10473961
• 发表时间: 2023-11
• 期刊: 2024 29th Asia and South Pacific Design Automation Conference (ASP-DAC)
• 作者: Zhuoping Yang;Shixin Ji;Xingzhen Chen;Jinming Zhuang;Weifeng Zhang;Dharmesh Jani;Peipei Zhou

SSR: Spatial Sequential Hybrid Architecture for Latency Throughput Tradeoff in Transformer Acceleration

• DOI: 10.1145/3626202.3637569
• 发表时间: 2024-01
• 期刊: Proceedings of the 2024 ACM/SIGDA International Symposium on Field Programmable Gate Arrays
• 作者: Jinming Zhuang;Zhuoping Yang;Shixin Ji;Heng Huang;Alex K. Jones;Jingtong Hu;Yiyu Shi;Peipei Zhou

AIM: Accelerating Arbitrary-Precision Integer Multiplication on Heterogeneous Reconfigurable Computing Platform Versal ACAP

目的：在异构可重构计算平台 Versal ACAP 上加速任意精度整数乘法

• DOI: 10.1109/iccad57390.2023.10323754
• 发表时间: 2023
• 期刊: IEEE
• 作者: Yang, Zhuoping;Zhuang, Jinming;Yin, Jiaqi;Yu, Cunxi;Jones, Alex K.;Zhou, Peipei
• 通讯作者: Zhou, Peipei

High Performance, Low Power Matrix Multiply Design on ACAP: from Architecture, Design Challenges and DSE Perspectives

ACAP 上的高性能、低功耗矩阵乘法设计：来自架构、设计挑战和 DSE 角度

• DOI: 10.1109/dac56929.2023.10247981
• 发表时间: 2023
• 期刊: 2023 60th ACM/IEEE Design Automation Conference (DAC
• 作者: Zhuang, Jinming;Yang, Zhuoping;Zhou, Peipei
• 通讯作者: Zhou, Peipei

CHARM: Composing Heterogeneous AcceleRators for Matrix Multiply on Versal ACAP Architecture

CHARM：在 Versal ACAP 架构上组合用于矩阵乘法的异构加速器

• DOI: 10.1145/3543622.3573210
• 发表时间: 2023
• 期刊: Proceedings of the 2023 ACM/SIGDA International Symposium on Field Programmable Gate Arrays
• 作者: Zhuang, Jinming;Lau, Jason;Ye, Hanchen;Yang, Zhuoping;Du, Yubo;Lo, Jack;Denolf, Kristof;Neuendorffer, Stephen;Jones, Alex;Hu, Jingtong
• 通讯作者: Hu, Jingtong