Elements: Development of Assumption-Free Parallel Data Curing Service for Robust Machine Learning and Statistical Predictions

要素：开发用于鲁棒机器学习和统计预测的无假设并行数据固化服务

• 批准号: 1931380
• 负责人: In Ho Cho
• 金额: $ 59.24万
• 依托单位: Iowa State University
• 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: 2019-09-01 至 2023-11-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1931380&HistoricalAwards=false
• 关键词: Elements; Development; Assumption; Free; Parallel

Large, incomplete datasets create major challenges for statistical prediction in research. This project will develop a data curing service that is able to manage large, incomplete, and diverse datasets, and would provide uncertainty measures for the cured data. The project identifies and collaborates with several communities where this data service is central to scientific research, including civil engineering, building science, urban energy, and social science. The effort creates a parallel data curing service, provides uncertainty measures for the cured data, and develops supplementary imputing algorithms. The team develops a data curing platform with imputation for incomplete, heterogeneous data; robust machine learning (ML) and statistical predictions would be established by developing an easy-to-use, general-purpose, large data-friendly imputation program. The focus is on a novel combination of three established imputation methods: two-level finite mixture model-based imputation (FMMI), fractional hot deck imputation (FHDI), and Gaussian mixture model-based imputation (GMMI), for which parallel implementations in R would also be provided. This award by the NSF Office of Advanced Cyberinfrastructure is jointly funded by the Established Program to Stimulate Competitive Research (EPSCoR).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.

庞大的、不完整的数据集给研究中的统计预测带来了重大挑战。该项目将开发一种数据修复服务，能够管理大型、不完整和多样化的数据集，并将为修复的数据提供不确定性测量。该项目确定并与几个社区合作，在这些社区中，这种数据服务是科学研究的核心，包括土木工程、建筑科学、城市能源和社会科学。这项工作创建了一个并行的数据修复服务，为修复的数据提供了不确定性度量，并开发了补充的输入算法。该团队开发了一个数据固化平台，可以对不完整的、不同种类的数据进行计算；通过开发一个易于使用、通用的大型数据友好计算程序，将建立稳健的机器学习(ML)和统计预测。重点研究了基于两级有限混合模型的推算(FMMI)、基于分数热甲板推算(FHDI)和基于高斯混合模型的推算(GMMI)这三种推算方法的新组合，并提供了R语言中的并行实现。这一奖项由NSF高级网络基础设施办公室联合资助，由既定的激励竞争研究计划(EPSCoR)资助。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势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

Statistical inference using regularized M-estimation in the reproducing kernel Hilbert space for handling missing data

• DOI: 10.1007/s10463-023-00872-8
• 发表时间: 2021-07
• 期刊: Annals of the Institute of Statistical Mathematics
• 影响因子: 1
• 作者: Hengfang Wang;Jae Kwang Kim

Statistical inference with semiparametric nonignorable nonresponse models

• DOI: 10.1111/sjos.12652
• 发表时间: 2023-04
• 期刊: Scandinavian Journal of Statistics
• 影响因子: 1
• 作者: Masatoshi Uehara;Danhyang Lee;Jae Kwang Kim

A framework for glass-box physics rule learner and its application to nano-scale phenomena

• DOI: 10.1038/s42005-020-0339-x
• 发表时间: 2020-05-08
• 期刊: COMMUNICATIONS PHYSICS
• 影响因子: 5.5
• 作者: Cho, In Ho;Li, Qiang;Kim, Jaeyoun
• 通讯作者: Kim, Jaeyoun

Survey data integration for regression analysis using model calibration.

• 发表时间: 2021-07
• 期刊: arXiv: Methodology
• 作者: Zhonglei Wang;Hang J Kim;Jae Kwang Kim

Flexible and interpretable generalization of self-evolving computational materials framework

• DOI: 10.1016/j.compstruc.2021.106706
• 发表时间: 2021-11-17
• 期刊: COMPUTERS & STRUCTURES
• 影响因子: 4.7
• 作者: Bazroun, Mohammed;Yang, Yicheng;Cho, In Ho
• 通讯作者: Cho, In Ho