Collaborative Research: CISE-MSI: RPEP: III: celtSTEM Research Collaborative: Catapulting MSI Faculty and Students into Computational Research.

合作研究：CISE-MSI：RPEP：III：celtSTEM 研究合作：将 MSI 教师和学生推向计算研究。

• 批准号: 2131294
• 负责人: Christopher Jermaine
• 金额: $ 48.85万
• 依托单位: William Marsh Rice University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-11-01 至 2024-10-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2131294&HistoricalAwards=false
• 关键词: Collaborative; Research; CISE; MSI; RPEP

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).As datasets have grown in size and complexity, the data they contain are essential for advancing scientific discoveries. Computational methods such as machine learning and data mining play a key role in facilitating the analysis of large datasets. However, the tools used to store and manage datasets and the tools used for machine learning are largely separate and treat data differently. Machine learning algorithms for data that has a network structure, in which a key feature of the data is relationships between items in the network, are also less well-developed compared to machine learning algorithms that treat items independently. This project will advance network-based machine learning algorithms, developing new approaches that are well-matched to common database technologies and apply them to solve foundational problems in biology including genome and proteome sequences. The project will also develop a long-term computational research initiative at the lead institution, which is a Minority Serving Institution (MSI), through working with a research-intensive organization along with industry and government partners, to implement strategies for preparing MSI students and faculty to excel in computational research.The machine learning methods portion of the project will scale neural computations to huge graphs using graph neural networks. Current approaches are limited because of problems scaling to very large graphs and effectively distributing computations across multiple machines. The approach focuses on leveraging relational database technologies, which are ideally situated to overcome these limitations due to the close link between graphs and relations and the rich set of tools they already possess for optimizing computation across large datasets. The tools developed will be used to help solve foundational problems in biology, focusing on two projects. The first involves machine learning-aided search of large protein databases for unique sequence patterns, developing new high-dimensional data encodings and graph-based algorithms to facilitate evolutionary, structural, functional, and ontological searches. The second involves analysis of metagenomic data to identify viruses (and variants) carried by mosquitos, developing novel graph encodings of mosquitos’ DNA and RNA along with machine learning-based analyses of them to analyze nucleic acid sequence data and expand the understanding of the viral collections carried by insects that interact closely with human populations. The research will be carried out in close collaboration between the lead university partners, who will develop coursework, team structures and practices, and mentoring approaches that provide a framework in which MSI students and faculty can gain both research skills and opportunities. This capacity-building work will be guided by a situated learning pedagogy and a socio-technical lens that emphasizes context and relationships between people and technology.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.

该奖项全部或部分由2021年美国救援计划法案（公法117-2）资助。随着数据集的规模和复杂性不断增长，它们所包含的数据对于推进科学发现至关重要。机器学习和数据挖掘等计算方法在促进大型数据集的分析方面发挥着关键作用。然而，用于存储和管理数据集的工具和用于机器学习的工具在很大程度上是分开的，并且以不同的方式处理数据。对于具有网络结构的数据，其中数据的关键特征是网络中项目之间的关系，与独立处理项目的机器学习算法相比，机器学习算法的发展也不太好。该项目将推进基于网络的机器学习算法，开发与常见数据库技术相匹配的新方法，并将其应用于解决生物学中的基础问题，包括基因组和蛋白质组序列。该项目还将通过与研究密集型组织沿着与行业和政府合作伙伴合作，在牵头机构（少数民族服务机构（MSI））制定长期计算研究计划，实施战略，为MSI的学生和教师准备在计算研究中脱颖而出。该项目的机器学习方法部分将使用图神经网络将神经计算扩展到巨大的图网络.目前的方法是有限的，因为问题扩展到非常大的图形和有效地分布在多台机器上的计算。该方法的重点是利用关系数据库技术，这些技术非常适合克服这些限制，因为图形和关系之间的密切联系以及它们已经拥有的用于优化大型数据集计算的丰富工具集。开发的工具将用于帮助解决生物学中的基础问题，重点是两个项目。第一个涉及机器学习辅助搜索大型蛋白质数据库中的独特序列模式，开发新的高维数据编码和基于图形的算法，以促进进化，结构，功能和本体搜索。第二个涉及分析宏基因组数据以识别蚊子携带的病毒（和变体），开发蚊子DNA和RNA的新型图形编码，沿着基于机器学习的分析，以分析核酸序列数据，并扩大对与人类密切互动的昆虫携带的病毒集合的理解。该研究将在领先的大学合作伙伴，谁将开发课程，团队结构和实践之间的密切合作进行，并提供一个框架，其中MSI的学生和教师可以获得研究技能和机会的指导方法。这项能力建设工作将以情境学习教学法和社会技术透镜为指导，强调人与技术之间的背景和关系。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

Auto-Differentiation of Relational Computations for Very Large Scale Machine Learning

• DOI: 10.48550/arxiv.2306.00088
• 发表时间: 2023-05
• 作者: Yu-Shuen Tang;Zhimin Ding;Dimitrije Jankov;Binhang Yuan;Daniel Bourgeois;C. Jermaine