HDR TRIPODS: Penn Institute for Foundations of Data Science

HDR TRIPODS：宾夕法尼亚大学数据科学研究所

• 批准号: 1934876
• 负责人: Shivani Agarwal
• 金额: $ 131.05万
• 依托单位: University of Pennsylvania
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1934876&HistoricalAwards=false
• 关键词: HDR; TRIPODS; Penn; Institute; Foundations

The growing field of data science promises to bring many benefits to society: personalized knowledge and services; improved healthcare; improved decision-making at individual, organizational, national, and international levels; a safer and possibly fairer society; and many others. The ability to realize these promises, however, depends critically on building the right foundational principles for the field. This project establishes an NSF TRIPODS Institute, termed the Penn Institute for Foundations of Data Science (PIFODS), at the University of Pennsylvania, with the goal of bringing together scientists and ideas from multiple disciplines, including computer science, electrical engineering, statistics, and mathematics, in order to collectively develop long-lasting principles for data science that can serve the field for decades to come. The main activities of the Institute will include transdisciplinary research, education and training, engagement with the broader research community through invited seminars and workshops, and engagement with applied scientists and practitioners. The PIFODS team seeks to develop principles for the following five thrusts: principles for complex learning tasks; principles for efficient optimization (convex, non-convex, and submodular); principles for streaming, distributed, and massively parallel data analysis; principles for privacy-preserving and fairness-preserving data analysis; and principles for reproducible data analysis. Each of these thrusts addresses an important foundational need in data science. These needs range from designing learning algorithms with stronger performance guarantees, and developing principles for optimization in adaptive settings, to developing a fundamental understanding of the tradeoffs between various modern computational resources in data science, as well as developing data science algorithms that guarantee meaningful notions of privacy, fairness, and reproducibility. Each thrust requires interactions among several of the TRIPODS disciplines; several of these thrusts also naturally interact with each other. On the education and training side, the PIFODS team has already initiated several new transdisciplinary courses related to data science that are aimed at developing a common language across disciplines; under the aegis of the Institute, the team will continue to further develop and refine these courses, and will incorporate feedback from these courses to inform the university's emerging transdisciplinary data science curriculum. On the applications side, the PIFODS team will actively engage with applied scientists and practitioners of data science, including both members of the broader university community and selected industry practitioners; these engagements will both help to inform possible additional research thrusts in the future, and help to solve important data-driven problems in society. This project is part of the National Science Foundation's Harnessing the Data Revolution (HDR) Big Idea activity.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.

不断发展的数据科学领域有望为社会带来许多好处：个性化的知识和服务；改善医疗保健；改进个人、组织、国家和国际层面的决策；一个更安全、可能更公平的社会；以及许多其他人。然而，实现这些承诺的能力关键取决于为该领域建立正确的基础原则。该项目在宾夕法尼亚大学建立了一个 NSF TRIPODS 研究所，称为宾夕法尼亚数据科学研究所 (PIFODS)，其目标是汇集来自多个学科的科学家和想法，包括计算机科学、电气工程、统计学和计算机科学。数学，以便共同开发数据科学的持久原理，可以在未来几十年内服务于该领域。该研究所的主要活动将包括跨学科研究、教育和培训、通过受邀研讨会和讲习班与更广泛的研究界接触，以及与应用科学家和从业者的接触。 PIFODS 团队力求为以下五个重点制定原则：复杂学习任务的原则；有效优化的原则（凸、非凸和子模）；流式、分布式和大规模并行数据分析的原理；保护隐私和公平的数据分析原则；以及可重复数据分析的原则。这些主旨中的每一个都满足了数据科学的重要基础需求。这些需求的范围包括设计具有更强性能保证的学习算法，开发自适应设置中的优化原则，发展对数据科学中各种现代计算资源之间的权衡的基本理解，以及开发保证有意义的概念的数据科学算法。隐私、公平和可重复性。每个推动力都需要 TRIPODS 多个学科之间的相互作用；其中一些推动力也自然地相互作用。在教育培训方面，PIFODS团队已经启动了几门与数据科学相关的新跨学科课程，旨在开发跨学科的共同语言；在该研究所的支持下，该团队将继续进一步开发和完善这些课程，并将吸收这些课程的反馈，为该大学新兴的跨学科数据科学课程提供信息。在应用方面，PIFODS团队将积极与应用科学家和数据科学从业者合作，包括更广泛的大学社区成员和选定的行业从业者；这些参与既有助于为未来可能的其他研究方向提供信息，也有助于解决社会中重要的数据驱动问题。该项目是美国国家科学基金会利用数据革命 (HDR) 大创意活动的一部分。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力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

Algorithms and Learning for Fair Portfolio Design

公平投资组合设计的算法和学习

• 发表时间: 2021-01
• 期刊: Economics and Computation (EC
• 作者: Diana, Emily;Dick, Travis;Elzayn, Hadi;Kearns, Michael;Roth, Aaron;Schutzman, Zachary;Sharifi;Ziani, Juba
• 通讯作者: Ziani, Juba

A snapshot of the frontiers of fairness in machine learning

机器学习公平前沿概览

• DOI: 10.1145/3376898
• 发表时间: 2020-04
• 期刊: Communications of the ACM
• 影响因子: 22.7
• 作者: Chouldechova, Alexandra;Roth, Aaron
• 通讯作者: Roth, Aaron

Online Multivalid Learning: Means, Moments, and Prediction Intervals

在线多重学习：均值、矩和预测区间

• 发表时间: 2022-06
• 期刊: {ITCS}
• 作者: Varun Gupta; Christopher Jung
• 通讯作者: Christopher Jung

Near-linear Size Hypergraph Cut Sparsifiers

近线性尺寸超图切割稀疏器

• DOI: 10.1109/focs46700.2020.00015
• 发表时间: 2020-11
• 期刊: 61st IEEE Annual Symposium on Foundations of Computer Science
• 作者: Chen, Yu;Khanna, Sanjeev;Nagda, Ansh
• 通讯作者: Nagda, Ansh

Convex calibrated surrogates for the multi-label F-measure

多标签 F 测量的凸校准替代项

• 发表时间: 2020-01
• 期刊: International conference on machine learning
• 作者: Zhang, M;Ramaswamy, H;Agarwal, S
• 通讯作者: Agarwal, S