NSF Convergence Accelerator Track D: A Trusted Integrative Model and Data Sharing Platform for Accelerating AI-Driven Health Innovation

NSF 融合加速器轨道 D：加速人工智能驱动的健康创新的可信集成模型和数据共享平台

• 批准号: 2040588
• 负责人: Hai Li
• 金额: $ 96.61万
• 依托单位: Duke University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-09-15 至 2022-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2040588&HistoricalAwards=false
• 关键词: NSF; Convergence; Accelerator; Track; Trusted

The NSF Convergence Accelerator supports use-inspired, team-based, multidisciplinary efforts that address challenges of national importance and will produce deliverables of value to society in the near future. This project, NSF Convergence Accelerator Track D: A Trusted Integrative Model and Data Sharing Platform for Accelerating AI-Driven Health Innovation, will develop a novel health-related federated learning and model-sharing platform, LEARNER, to enable collaborative big data mining for biomedical applications by integrating cross-disciplinary expertise from machine learning, trustworthy AI, and biomedical data science. LEARNER will incorporate novel asynchronous federated learning algorithms based on rigorous theoretical foundations using trustworthy AI techniques, fairness-aware and interpretable machine learning models, large-scale computational strategies and effective software tools to reveal the complex relationships among heterogeneous health data. The project will address critical challenges in exploiting big data for biomedical and health, which include access to large data collections, computational intensity of AI/ML algorithms, complexity of hyperparameter tuning, and the need for effective multidisciplinary expertise and collaboration. Data privacy is another critical concern since health data is intrinsically sensitive and could be exploited to reveal an individual’s identity even when the data are carefully anonymized. LEARNER will include a suite of collaborative data analysis and privacy-preserving mechanisms and tools that will securely support various types of health data analytics, including mechanisms to detect potential data privacy leakages. Machine learning models typically involve complex procedures for optimization and the induced results can be difficult to interpret, and to replicate and reproduce. Novel methods will be employed to improve the interpretability and reproducibility of complex health data analytics models.The project team, with individuals from academia and industry, will develop an interdisciplinary program for training and education of graduate and undergraduate students. A cross-disciplinary course will also be developed on Health Data Science for beginning graduate students and senior undergraduate students from a variety of programs, including Computer Science and Engineering, Informatics, Electrical Engineering, Biomedical Engineering, Biology, and Statistics. The project will put special emphasis on attracting female and under-represented minority students to explore advanced computational technologies in the context of the LEARNER platform. Interested senior undergraduate students will be able to work on well-defined and well-scoped small projects, which will enable them to work with graduate students and the PI team of the project. Such project could also be undertaken as summer projects by undergraduate students in science and engineering.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.

NSF融合加速器支持以团队为基础的多学科努力，这些努力应对国家重要性的挑战，并将在不久的将来为社会带来价值的可交付成果。该项目，NSF收敛加速器轨道D：一个可信赖的集成模型和用于加速AI驱动的健康创新的数据共享平台，将开发一个与健康相关的新型联邦学习和模型共享平台，以启用与机器人学习的跨学科研究的生物医学应用程序，以启用生物医学应用程序的协作大数据挖掘，这学习者将使用可信赖的AI技术，公平感和可解释的机器学习模型，大规模的计算策略和有效的软件工具来融合基于严格的理论基础的新型异步联合学习算法，以揭示异质健康数据之间的复杂关系。该项目将针对利用大数据进行生物医学和健康方面的关键挑战，包括访问大型数据收集，AI/ML算法的计算强度，超参数调整的复杂性以及有效的多学科专业知识和协作的需求。数据隐私是另一个关键问题，因为健康数据本质上是敏感的，并且可以探索以揭示个人的身份，即使数据仔细匿名化。学习者将包括一系列协作数据分析和隐私保护机制和工具，这些机制将安全支持各种类型的健康数据分析，包括检测潜在数据隐私泄漏的机制。机器学习模型通常涉及复杂的优化程序，并且诱导的结果可能难以解释，复制和再现。将聘请新颖的方法来改善复杂健康数据分析模型的可解释性和可重复性。项目团队将与来自学术界和行业的个人一起制定一项跨学科计划，用于培训和教育研究生和本科生。还将在健康数据科学上开发跨学科课程，旨在为初学者和来自各种课程的高级本科生，包括计算机科学和工程，信息学，电气工程，生物医学工程，生物学和统计学。该项目将特别强调吸引女性和代表性不足的少数民族学生在学习者平台的背景下探索先进的计算技术。有兴趣的高级本科生将能够从事定义明确且分布良好的小型项目，这将使他们能够与研究生和项目的PI团队合作。该项目也可以由科学和工程学的本科生作为夏季项目进行。该奖项反映了NSF的法定任务，并使用基金会的知识分子优点和更广泛的影响评估标准，被认为是通过评估而被视为珍贵的。

项目成果

Communication-Efficient Projection-Free Algorithm for Nonconvex Constrained Learning Models

非凸约束学习模型的通信高效无投影算法

• 发表时间: 2021
• 期刊: 35th AAAI Conference on Artificial Intelligence (AAAI 2021
• 作者: Wenhan Xian, Feihu Huang

A Faster Decentralized Algorithm for Nonconvex Minimax Problems

• 发表时间: 2021
• 作者: Wenhan Xian;Feihu Huang;Yanfu Zhang;Heng Huang

Step-Ahead Error Feedback for Distributed Training with Compressed Gradient

• DOI: 10.1609/aaai.v35i12.17254
• 发表时间: 2020-08
• 作者: An Xu;Zhouyuan Huo;Heng Huang

Closing the Generalization Gap of Cross-silo Federated Medical Image Segmentation

• DOI: 10.1109/cvpr52688.2022.02020
• 发表时间: 2022-03
• 期刊: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
• 作者: An Xu;Wenqi Li;Pengfei Guo;Dong Yang;H. Roth;Ali Hatamizadeh;Can Zhao;Daguang Xu;Heng Huang;Ziyue Xu-

Detached Error Feedback for Distributed SGD with Random Sparsification

• 发表时间: 2020-04
• 作者: An Xu;Heng Huang