SI2-SSE: Human- and Machine-Intelligent Software Elements for Cost-Effective Scientific Data Digitization

SI2-SSE：用于经济高效的科学数据数字化的人机智能软件元素

• 批准号: 1535086
• 负责人: Jose Fortes
• 金额: $ 48.8万
• 依托单位: University of Florida
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2015
• 资助国家: 美国
• 起止时间: 2015-08-01 至 2020-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1535086&HistoricalAwards=false
• 关键词: SI2; SSE; Human; Machine; Intelligent

In the era of data-intensive scientific discovery, Big Data scientists in all communities spend the majority of their time and effort collecting, integrating, curating, transforming, and assessing quality before actually performing discovery analysis. Some endeavors may even start from information not being available and accessible in digital form, and when it is available, it is often in non-structured form, not compatible with analytics tools that require structured and uniformly-formatted data. Two main methods to deal with the volume and variety of data as well as to accelerate the rate of digitization have been to apply crowdsourcing or machine-learning solutions. However, very little has been done to simultaneously take advantage of both types of solutions, and to make it easier for different efforts to share and reuse developed software elements. The vision of the Human- and Machine-Intelligent Network (HuMaIN) project is to accelerate scientific data digitization through fundamental advances in the integration and mutual cooperation between human and machine processing in order to handle practical hurdles and bottlenecks present in scientific data digitization. Even though HuMaIN concentrates on digitization tasks faced by the biodiversity community, the software elements being developed are generic in nature, and expected to be applicable to other scientific domains (e.g., exploring the surface of the moon for craters require the same type of crowdsourcing tool as finding words in text, and the same questions of whether machine-learning tools could provide similar results can be tested).The HuMaIN project proposes to conduct research and develop the following software elements: (a) configurable Machine-Learning applications for scientific data digitization (e.g., Optical Character Recognition and Natural Language Processing), which will be made automatically available as RESTful services for increasing the ability of HuMaIN software elements to interoperate with other elements while decreasing the software development time via a new application specification language; (b) workflows leading to a cyber-human coordination system that will take advantage of feedback loops (e.g., based on consensus of crowdsourced data and its quality) for self-adaptation to changes and increased sustainability of the overall system, (c) new crowdsourcing micro-tasks with ability of being reusable for a variety of scenarios and containing user activity sensors for studying time-effective user interfaces, and (d) services to support automated creation and configuration of crowdsourcing workflows on demand to fit the needs of individual groups. A cloud-based system will be deployed to provide the necessary execution environment with traceability of service executions involved in cyber-human workflows, and cost-effectiveness analysis of all the software elements developed in this project will provide assessment and evaluation of long standing what-if scenarios pertaining human- and machine-intelligent tasks. Crowdsourcing activities will attract a wide range of users with tasks that require low expertise, and at the same time it will expose volunteers to applied science and engineering, potentially attracting interest of K-12 teachers and students.

在数据密集型科学发现的时代，所有社区的大数据科学家都花费大部分时间和精力来收集，整合，策划，转换和评估质量，然后再实际执行发现分析。某些努力甚至可能是从无法以数字形式获得的信息开始的，并且在可用的情况下，通常以非结构性形式，与需要结构化且均匀形成的数据的分析工具不兼容。处理数据数量和多种数据以及加速数字化速率的两种主要方法是应用众包或机器学习解决方案。但是，几乎没有采取很少的工作来同时利用两种类型的解决方案，并且使不同的努力更容易共享和重用开发的软件元素。人类和机器智能网络（Humain）项目的愿景是通过人类和机器处理之间的整合和相互合作的基本进步来加速科学数据数字化，以处理科学数据数字化中存在的实用障碍和瓶颈。即使腐殖质集中于生物多样性社区面临的数字化任务，但开发的软件要素本质上是仿制的，预计将适用于其他科学领域（例如，探索山口山的月球表面（a）可配置的用于科学数据数字化的机器学习应用程序（例如，光学角色识别和自然语言处理），将自动提供作为宁静服务，以提高Humain软件元素与其他元素互动的能力，同时通过新的应用程序规范语言减少软件开发时间； （b）导致网络人类协调系统的工作流程，该系统将利用反馈循环（例如，基于众包数据及其质量的共识）来自我适应整体系统的变化和增加的可持续性，（c）新的众包微型任务，并能够为各种式的驾驶用户提供服务，并能够为各种式的用户练习，并培养各种情况。按需众包工作流程的自动创建和配置以满足各个组的需求。将部署一个基于云的系统，以提供必要的执行环境，并以网络人类工作流程涉及的服务执行的可追溯性，以及对本项目中开发的所有软件元素的成本效益分析，将提供对长期存在的人类和机器和机器和机器智能任务的评估和评估。众包活动将吸引众多需要专业知识的任务的用户，同时，它将使志愿者接触到应用科学和工程，从而吸引K-12教师和学生的兴趣。

项目成果

Cooperative Human-Machine Data Extraction from Biological Collections

• DOI: 10.1109/escience.2016.7870884
• 发表时间: 2016-01-01
• 期刊: PROCEEDINGS OF THE 2016 IEEE 12TH INTERNATIONAL CONFERENCE ON E-SCIENCE (E-SCIENCE)
• 作者: Alzuru, Icaro;Matsunaga, Andrea;Fortes, Jose A. B.
• 通讯作者: Fortes, Jose A. B.

SELFIE: Self-Aware Information Extraction from Digitized Biocollections

自拍：从数字化生物收藏中提取自我意识信息

• DOI: 10.1109/escience.2017.19
• 发表时间: 2017
• 期刊: New Zealand
• 作者: Alzuru, Icaro;Matsunaga, Andrea;Tsugawa, Mauricio;Fortes, Jose A.B.
• 通讯作者: Fortes, Jose A.B.

Quality-Aware Human-Machine Text Extraction for Biocollections using Ensembles of OCRs

使用 OCR 集成对生物样本进行质量感知人机文本提取

• DOI: 10.1109/escience.2019.00020
• 发表时间: 2019
• 期刊: USA
• 作者: Alzuru, Icaro;Stephens, Rhiannon;Matsunaga, Andrea;Tsugawa, Mauricio;Flemons, Paul;Fortes, Jose A.B.
• 通讯作者: Fortes, Jose A.B.

Task Design and Crowd Sentiment in Biocollections Information Extraction

生物馆藏信息提取中的任务设计和人群情绪

• DOI: 10.1109/cic.2017.00056
• 发表时间: 2017
• 期刊: CA USA
• 作者: Alzuru, Icaro;Matsunaga, Andrea;Tsugawa, Mauricio;Fortes, Jose A.B.
• 通讯作者: Fortes, Jose A.B.

Human-Machine Information Extraction Simulator for Biological Collections

生物标本人机信息提取模拟器

• DOI: 10.1109/bigdata47090.2019.9005601
• 发表时间: 2019
• 期刊: USA
• 作者: Alzuru, Icaro;Malladi, Aditi;Matsunaga, Andrea;Tsugawa, Mauricio;Jose A.B., Fortes
• 通讯作者: Jose A.B., Fortes