Collaborative Research: Framework: Software: CINES: A Scalable Cyberinfrastructure for Sustained Innovation in Network Engineering and Science

合作研究：框架：软件：CINES：用于网络工程和科学持续创新的可扩展网络基础设施

• 批准号: 1835598
• 负责人: Jurij Leskovec
• 金额: $ 54万
• 依托单位: Stanford University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-11-01 至 2024-03-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1835598&HistoricalAwards=false
• 关键词: Collaborative; Research; Framework; Software; CINES

Networks are ubiquitous and are a part of our common vocabulary. Network science and engineering has emerged as a formal field over the last twenty years and has seen explosive growth. Ideas from network science are central to companies such as Akamai, Twitter, Google, Facebook, and LinkedIn. The concepts have also been used to address fundamental problems in diverse fields (e.g., biology, economics, social sciences, psychology, power systems, telecommunications, public health and marketing), and are now part of most university curricula. Ideas and techniques from network science are widely used in making scientific progress in the disciplines mentioned above. Networks are now part of the public vocabulary, with news articles and magazines frequently using the term "networks" to refer to interconnected entities. Yet, resources for effective use of techniques from network science are largely dispersed and stand-alone, of small scale, home-grown for personal use, and/or do not cover the broad range of operations that need to be performed on networks. Compositions of these diverse capabilities are rare. Furthermore, many researchers who study networks are not computer scientists. As a result, they do not have easy access to computing and data resources; this creates a barrier for researchers. This project will develop a sophisticated cyberinfrastructure that brings together various resources to provide a unifying ecosystem for network science that is greater than the sum of its parts. The resulting cyberinfrastructure will benefit researchers and students from various disciplines by facilitating access to various tools for synthesizing and analyzing large networks, and by providing access points for contributors of new software and data. An important benefit of the system is that it can be readily used even by researchers who have no formal training in computer programming. The cyberinfrastructure resulting from this work will foster multi-disciplinary and multi-university research and teaching collaborations. As part of this project, comprehensive education and outreach programs will be launched by the participating institutions, spanning educators and K-12 students. These programs will include network science courses with students from minority and under-represented groups, and students at smaller institutions who do not have easy access to high performance computing resources.Resources for doing network science are largely dispersed and stand-alone (in silos of isolated tools), of small scale, or home-grown for personal use. What is needed is a cyberinfrastructure to bring together various resources, to provide a unifying ecosystem for network science that is greater than the sum of its parts. The primary goal of this proposal is to build self-sustaining cyberinfrastructure (CI) named CINES (Cyberinfrastructure for Sustained Innovation in Network Engineering and Science) that will be a community resource for network science. CINES will be an extensible and sustainable platform for producers and consumers of network science data, information, and software. CINES will have: (1) a layered architecture that systematically modularizes and isolates messaging, infrastructure services, common services, a digital library, and APIs for change-out and updates; (2) a robust and reliable infrastructure that---for applications (apps)---is designed to accommodate technological advances in methods, programming languages, and computing models; (3) a resource manager to enable jobs to run on target machines for which they are best suited; (4) an engine to enable users to create new workflows by composing available components and to distribute the resulting workload across computing resources; (5) orchestration among system components to provide CI-as-a-service (CIaaS) that scales under high system load to networks with a billion or more vertices; (6) a digital library with 100,000+ networks of various kinds that allows rich services for storing, searching, annotating, and browsing; (7) structural methods (e.g., centrality, paths, cuts, etc.) and dynamical models of various contagion processes; (8) new methods to acquire data, build networks, and augment them using machine learning techniques; (9) a suite of industry- recognized tools such as SNAP, NetworkX, and R-studio that make it easier for researchers, educators, and analysts to do network science and engineering; (10) a suite of APIs that allows developers to add new web-apps and services, based on an app-store model, and allows access to CINES from third party software; and (11) metrics and a Stack Overflow model, among other features, for producers and consumers to interact (in real-time) and guide the evolution of CINES. CINES will enable fundamental changes in the way researchers study and teach complex networks. The use of state-of-the-art high-performance computing (HPC) resources to synthesize, analyze, and reason about large networks will enable researchers and educators to study networks in novel ways. CINES will allow scientists to address fundamental scientific questions---e.g., biologists can use network methods to reason about genomics data that is now available in large quantities due to fast and effective sequencing and the NIH Microbiome Program. It will enable educators to harness HPC technologies to teach Network Science to students spanning various academic levels, disciplines, and institutions. CINES, which will be useful to researchers supported by many NSF directorates and divisions, will be designed for scalability, usability, extensibility, and sustainability. This project will also advance the fields of digital libraries and cloud computing by stretching them to address challenges related to Network Science. Given the multidisciplinary nature of the field, CINES will provide a collaborative space for scientists from different disciplines, leading to important cross fertilization of ideas.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.

网络无处不在，是我们常用词汇的一部分。在过去的二十年里，网络科学与工程已经成为一个正式的领域，并出现了爆炸式的增长。网络科学的思想是Akamai、Twitter、b谷歌、Facebook和LinkedIn等公司的核心。这些概念也被用于解决不同领域的基本问题（例如，生物学、经济学、社会科学、心理学、电力系统、电信、公共卫生和市场营销），现在是大多数大学课程的一部分。网络科学的思想和技术被广泛应用于上述学科的科学进步。网络现在是公共词汇的一部分，新闻文章和杂志经常使用“网络”一词来指代相互联系的实体。然而，有效利用网络科学技术的资源在很大程度上是分散和独立的、小规模的、自产自用的，并且（或者）不包括需要在网络上执行的广泛操作。这些不同能力的组合是罕见的。此外，许多研究网络的研究人员并不是计算机科学家。因此，他们不容易获得计算和数据资源；这给研究人员制造了一个障碍。该项目将开发一个复杂的网络基础设施，汇集各种资源，为网络科学提供一个统一的生态系统，其效果大于各部分的总和。由此产生的网络基础设施将使来自不同学科的研究人员和学生受益，通过促进对综合和分析大型网络的各种工具的访问，并通过为新软件和数据的贡献者提供接入点。该系统的一个重要优点是，即使没有受过正规计算机编程训练的研究人员也可以很容易地使用它。这项工作产生的网络基础设施将促进多学科和多大学的研究和教学合作。作为该项目的一部分，参与机构将推出全面的教育和推广计划，涵盖教育工作者和K-12学生。这些项目将包括网络科学课程，学生来自少数民族和代表性不足的群体，以及小型机构的学生，他们不容易获得高性能计算资源。用于网络科学研究的资源大多是分散的、独立的（在孤立的工具筒仓中）、小规模的或自产自用的。我们需要的是一个网络基础设施，将各种资源汇集在一起，为网络科学提供一个统一的生态系统，这个生态系统大于各部分之和。该提案的主要目标是建立自我维持的网络基础设施（CI），名为CINES（网络工程和科学持续创新的网络基础设施），这将成为网络科学的社区资源。CINES将为网络科学数据、信息和软件的生产者和消费者提供可扩展和可持续的平台。CINES将具有：(1)分层架构，系统地模块化和隔离消息传递、基础设施服务、公共服务、数字图书馆和用于更改和更新的api；(2)稳健可靠的基础设施，用于应用程序（app），旨在适应方法、编程语言和计算模型方面的技术进步；(3)一个资源管理器，使作业能够在最适合它们的目标机器上运行；(4)一个引擎，使用户能够通过组合可用组件来创建新的工作流，并将由此产生的工作量分配到不同的计算资源；(5)在系统组件之间进行编排，以提供在高系统负载下可扩展到具有十亿或更多顶点的网络的ci即服务（CIaaS）；(6)拥有10万多个网络的数字图书馆，提供丰富的存储、搜索、注释和浏览服务；(7)结构方法（如中心性、路径、切口等）和各种传染过程的动力学模型；(8)获取数据、构建网络并使用机器学习技术增强网络的新方法；(9)一套行业公认的工具，如SNAP、NetworkX和R-studio，使研究人员、教育工作者和分析人员更容易进行网络科学和工程；（10）一套api，允许开发者基于应用商店模式添加新的网络应用程序和服务，并允许从第三方软件访问CINES；（11）指标和堆栈溢出模型，以及其他功能，供生产者和消费者（实时）交互并指导CINES的发展。CINES将使研究人员研究和教授复杂网络的方式发生根本性的变化。使用最先进的高性能计算（HPC）资源对大型网络进行综合、分析和推理，将使研究人员和教育工作者能够以新颖的方式研究网络。CINES将使科学家能够解决基本的科学问题。，生物学家可以使用网络方法来推断基因组学数据，由于快速有效的测序和美国国立卫生研究院微生物组计划，这些数据现在大量可用。它将使教育工作者能够利用高性能计算技术向不同学术水平、学科和机构的学生教授网络科学。CINES将被设计为可扩展性、可用性、可扩展性和可持续性，它将对许多NSF理事会和部门支持的研究人员有用。该项目还将通过扩展数字图书馆和云计算领域来解决与网络科学相关的挑战，从而推进数字图书馆和云计算领域。鉴于该领域的多学科性质，CINES将为来自不同学科的科学家提供一个合作空间，从而促进重要的思想交叉交流。该奖项反映了美国国家科学基金会的法定使命，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

GreaseLM: Graph REASoning Enhanced Language Models for Question Answering

GreaseLM：用于问答的图形推理增强语言模型

• DOI: 10.48550/arxiv.2201.08860
• 发表时间: 2022
• 期刊: International Conference on Representation Learning (ICLR
• 作者: Zhang, X;Bosselut, A;Yasunaga, M;Ren, H;Liang, P;Manning, C;Leskovec, J
• 通讯作者: Leskovec, J

Open-World Semi-Supervised Learning

• 发表时间: 2021-02
• 期刊: ArXiv
• 作者: Kaidi Cao;Maria Brbic;J. Leskovec

Data-Driven Real-Time Strategic Placement of Mobile Vaccine Distribution Sites

• DOI: 10.1101/2021.12.15.21267736
• 发表时间: 2021-12
• 作者: Z. Mehrab;M. Wilson;S. Chang;G. Harrison;B. Lewis;A. Telionis;J. Crow;D. Kim;S. Spillmann;K. Peters;J. Leskovec;M. Marathe

Identity-aware Graph Neural Networks

• DOI: 10.1609/aaai.v35i12.17283
• 发表时间: 2021-01
• 作者: Jiaxuan You;Jonathan M. Gomes-Selman;Rex Ying;J. Leskovec

Coresets for Data-efficient Training of Machine Learning Models

• 发表时间: 2019-06
• 作者: Baharan Mirzasoleiman;J. Bilmes;J. Leskovec