BIGDATA: IA: A multi-level approach for global optimization of the surveillance and control of infectious disease in the swine industry

大数据：IA：全球优化养猪业传染病监测和控制的多层次方法

• 批准号: 1838207
• 负责人: Beatriz Martinez Lopez
• 金额: $ 160万
• 依托单位: University of California-Davis
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1838207&HistoricalAwards=false
• 关键词: BIGDATA; IA; multi; level; approach

The United States livestock industry has an enormous socio-economic impact contributing to annual sales of $180 billion and 550,000 direct jobs. Its sustainability and success rely on the maintenance of good livestock health, high productivity, and efficiency. This requires effective analytical methods, prediction models, and decision tools. While a vast amount of data has been collected in all production processes, the integration and usage of such data to better inform decisions in livestock health has remained circumstantial. It is usually restricted to simple descriptive statistics or molecular analyses for specific aspects of animal breeding and pathogen diagnostics. The goal of this project is to develop a new, multi-scale, approach to bridge the gap between the data availability and its effective usage. The project will focus on the swine industry and its most economically devastating disease, the Porcine Reproductive and Respiratory Syndrome (PRRS). This will not only have a direct beneficial impact in the swine industry but will contribute to the better manage other livestock health problems, saving producers and US livestock industry millions of dollars yearly. This is the first principled decision framework for the livestock industry that integrates multi-level data to predict disease dynamics, detect changes in farm status, and optimize the use of testing, treatment and vaccination strategies. As a consequence, it is expected to improve animal health and welfare and secure the sustainability of US agriculture and food systems by providing a data-driven decision framework and tools that push the frontier of precision epidemiology. The outcome of this project will be widely disseminated through our education and extension program (BIGDATA-4- HEALTH) and the integration of methods in the Disease BioPortal platform. PIs have well integrated the research and education programs and will continue to do so. The project will generate new curriculum for multiple classes in computer science, animal science and veterinary science and will involve undergraduate and graduate students.The project proposes a principled data-driven decision framework for systematic PRRS prevention and control, based on the multi-level data sources collected during swine production, using novel data mining and machine learning techniques. The objectives are: 1) Early detection through efficient testing using a proactive and cost sensitive testing framework. 2) Systematic PRRS prevention and control by effective integration of testing, vaccination, and biosecurity implementation at a production system level. 3) Real- world experimentation and algorithm evaluation using both large-scale numerical evaluation with real traces and small-scale experimentations and real-time validation in five demonstration swine operations. 4) Education and extension program (BIGDATA-4-HEALTH) with training materials and technology transfer activities and the expansion of our user-friendly Disease BioPortal platform to facilitate the use of the developed methods to all industry stakeholders, researchers, and the general public. The intellectual merits of this proposal are multi-folds: the research team will develop novel mechanisms that advance the state-of-the-art in data-driven decision making algorithms. To reduce exploration cost, efficient situation-aware exploration techniques that addresses the fundamental exploration-exploitation tradeoff in multi-armed bandits and reinforcement learning will be developed, To handle missing data, novel compressive sensing algorithms designed to better manage accuracy disparity will be investigated. Finally, to balance the cost-accuracy tradeoff, efficient means to integrate simulation and experimentation will be explored, which only has been limited studied in the literature. Furthermore, the value of these tools for the early detection of diseases at farm and system level using historical data and prospective real world experimentation will be also evaluated and demonstrated. While focusing on swine production, this work provides the foundations and can be adapted to improve animal health of other livestock species and to advance in other disciplines facing the same data challenges.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.

美国畜牧业具有巨大的社会经济影响力，年销售额达1 800亿美元，直接创造了55万个就业机会。它的可持续性和成功依赖于保持良好的牲畜健康，高生产力和效率。这需要有效的分析方法、预测模型和决策工具。虽然在所有生产过程中收集了大量数据，但整合和使用这些数据以更好地为牲畜健康决策提供信息仍然是偶然的。它通常限于简单的描述性统计或分子分析，用于动物育种和病原体诊断的特定方面。该项目的目标是开发一种新的、多尺度的方法，以弥合数据可用性与其有效使用之间的差距。该项目将重点关注养猪业及其最具经济破坏性的疾病，即猪繁殖与呼吸综合征（PRRS）。这不仅将对养猪业产生直接的有益影响，而且将有助于更好地管理其他牲畜健康问题，每年为生产者和美国畜牧业节省数百万美元。这是畜牧业的第一个原则性决策框架，它整合了多层次数据，以预测疾病动态，检测农场状态的变化，并优化检测，治疗和疫苗接种策略的使用。因此，预计它将通过提供数据驱动的决策框架和工具来改善动物健康和福利，并确保美国农业和食品系统的可持续性，从而推动精确流行病学的前沿。该项目的成果将通过我们的教育和推广计划（BIGDATA-4- HEALTH）以及疾病BioPortal平台中方法的整合进行广泛传播。PI已经很好地整合了研究和教育计划，并将继续这样做。该项目将为计算机科学、动物科学和兽医科学的多个课程提供新的课程，并将涉及本科生和研究生。该项目提出了一个原则性的数据驱动的决策框架，用于系统的PRRS预防和控制，基于在养猪生产过程中收集的多层次数据源，使用新的数据挖掘和机器学习技术。目标是：1）通过使用主动和成本敏感的测试框架进行有效的测试来早期发现。2)通过在生产系统层面有效整合检测、疫苗接种和生物安全实施，系统地预防和控制PRRS。3)真实的-使用具有真实的轨迹的大规模数值评估和小规模实验以及在五个示范猪操作中的实时验证进行的实验和算法评估。4)教育和推广计划（BIGDATA-4-HEALTH），包括培训材料和技术转让活动，以及扩展我们的用户友好型疾病BioPortal平台，以促进所有行业利益相关者，研究人员和公众使用开发的方法。这一提议的智力价值是多方面的：研究团队将开发新的机制，推动数据驱动决策算法的发展。为了降低探索成本，将开发有效的情境感知探索技术，解决多臂强盗和强化学习中的基本探索-利用权衡，为了处理丢失的数据，将研究旨在更好地管理精度差异的新型压缩感知算法。最后，为了平衡成本和精度的权衡，将探索有效的手段来集成仿真和实验，这在文献中只进行了有限的研究。此外，还将评估和证明这些工具在农场和系统层面使用历史数据和前瞻性真实的世界实验进行疾病早期检测的价值。这项工作的重点是养猪业，同时也为改善其他牲畜品种的动物健康以及在面临相同数据挑战的其他学科中取得进展奠定了基础。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

A Target-Agnostic Attack on Deep Models: Exploiting Security Vulnerabilities of Transfer Learning

• 发表时间: 2019-04
• 期刊: ArXiv
• 作者: Shahbaz Rezaei;Xin Liu

CTS2: Time Series Smoothing with Constrained Reinforcement Learning

• 发表时间: 2021
• 期刊: IEEE Robotics and Automation Letters
• 影响因子: 5.2
• 作者: Yongshuai Liu;Xin Liu;I. Tsang;X. Liu;Liu Liu-Liu

Exploiting Unlabeled Data to Improve Detection of Visual Anomalies in Soft Fruits

利用未标记的数据改进软水果视觉异常的检测

• 发表时间: 2022
• 期刊: AAAI-2023 Workshop on AI for Agriculture and Food Systems
• 作者: Choi T, Liu X

Impact of sensor data pre-processing strategies and selection of machine learning algorithm on the prediction of metritis events in dairy cattle

传感器数据预处理策略和机器学习算法选择对奶牛子宫炎事件预测的影响

• DOI: 10.1016/j.prevetmed.2023.105903
• 发表时间: 2023
• 期刊: Preventive Veterinary Medicine
• 影响因子: 2.6
• 作者: Vidal, Gema;Sharpnack, James;Pinedo, Pablo;Tsai, I Ching;Lee, Amanda Renee;Martínez-López, Beatriz
• 通讯作者: Martínez-López, Beatriz

On the Difficulty of Membership Inference Attacks

• DOI: 10.1109/cvpr46437.2021.00780
• 发表时间: 2021-06
• 期刊: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
• 作者: Shahbaz Rezaei;Xin Liu