HDR TRIPODS: Collaborative Research: Foundations of Greater Data Science

罗切斯特大学和康奈尔大学共同成立了大数据科学合作研究所(GDSC)。GDSC基于两个创始原则。首先，数据科学的持久发展需要结合电气工程、数学、统计学和理论计算机科学的技术和观点。研究人员的目标是在数据科学上形成一种超越任何个人领域的共识观点。第二，数据科学研究必须立足于应用领域。这有助于确保对数据可用性和质量的假设是现实的，并允许从实验和理论上检验方法结果。因此，GDSC的目标是考虑在医药和医疗保健领域的应用，这是一个重要的应用领域，数据科学的进步可以对社会产生直接的积极影响。GDSC的目标是解决由医疗保健问题引发的基本问题，获得将领域专业知识与与应用无关的方法相融合的解决方案，并最终产生影响医疗保健提供方式的科学进步。GDSC的目标是利用两个机构的地理上的接近，以及上述每个核心学科和医学领域的独特优势。GDSC的跨学科研究方向包括：(I)拓扑数据分析。高维、不完整和噪声数据带来的挑战是巨大的，但在许多应用中，利用问题的拓扑本质是可能的。GDSC的目标是开发新的基本方法和理论，以严格探索这种独特方法的前景。(Ii)数据表示。数据压缩、嵌入和降维在数据科学中扮演着重要的角色。受到生物医学成像、基因组学和神经棘波训练数据的新核心挑战的启发，GDSC的目标是开发新的源模型和失真测量，并最终寻求跨领域和学科的统一理论框架。(Iii)网络和图表学习。将数据科学应用于非同质群体的许多基本挑战最好通过网络或图表结构来探索。GDSC的目标是开发新的技术来解决谱社区检测中的参数依赖特征值问题，网络上的密度估计方法，以及研究时变网络中动态变量关系的时变图形模型的理论框架。(4)决策、控制和动态学习。循序渐进的决定在医学上事关重大。GDSC的目标是利用系统和控制工程方法来改善健康和疾病管理，并为标签有效的主动学习和动态治疗制度开发新的基础理论和方法。(五)多样和复杂的方式。大数据是复杂的数据，需要重大的新创新。GDSC的目标是开发在计算和隐私约束下的推理理论框架，以及在没有参数模型假设的情况下用于高维数据的推理。文本、图像和音频数据带来了进一步的挑战。为了应对这些挑战，GDSC旨在探索用于自然语言图形分析的转换系统和用于完全多模式分析的新的融合方法。该项目是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优势和更广泛的影响审查标准进行评估，被认为值得支持。

The University of Rochester and Cornell University jointly establish the Greater Data Science Cooperative Institute (GDSC). The GDSC is based on two founding tenets. The first is that enduring advances in data science require combining techniques and viewpoints across electrical engineering, mathematics, statistics, and theoretical computer science. The investigators' goal is to forge a consensus perspective on data science that transcends any individual field. The second is that data-science research must be grounded in an application domain. This helps to ensure that assumptions about the availability and quality of data are realistic, and it allows methodological results to be tested experimentally as well as theoretically. As such, the GDSC aims to consider applications in medicine and healthcare, an important application domain and one for which advances in data science can have a direct, positive impact on society. The GDSC aims to tackle foundational questions that are motivated by problems in healthcare, obtain solutions that fuse domain expertise with application-agnostic methodologies, and ultimately yield scientific advances that impact the way healthcare is provided. The GDSC aims to leverage the physical proximity of the two institutions, and the unique strengths in each of the core disciplines above and in medicine.The GDSC's cross-disciplinary research directions include: (i) Topological Data Analysis. The challenges that high-dimensional, incomplete, and noisy data present are great, but in many applications, exploiting the topological nature of the problem is possible. GDSC aims to develop new fundamental methods and theory to rigorously explore the promise of this unique approach. (ii) Data Representation. Data compression, embeddings, and dimension reduction play a fundamental role in data science. Inspired by new core challenges in biomedical imaging, genomics, and neural-spike training data, GDSC aims to develop novel source models and distortion measures, and ultimately seek a unifying theoretical framework across domains and disciplines. (iii) Network & Graph Learning. Many of the fundamental challenges in applying data science to non-homogeneous populations are best explored through a network or graph structure. GDSC aims to develop new techniques for parameter-dependent eigenvalue problems in spectral community detection, density-estimation methods on networks, and a theoretical framework for time-varying graphical models to study dynamic variable relations in time-evolving networks. (iv) Decisions, Control & Dynamic Learning. Sequential decisions are high-stakes in medicine. GDSC aims to utilize systems and control-engineering methods to improve health and disease management and develop new foundational theories and methods for label-efficient active learning and dynamic treatment regimes. (v) Diverse & Complex Modalities. Big data is complex data, and major new innovations are needed. GDSC aims to develop theoretical frameworks for inference under computational and privacy constraints and for high-dimensional data without parametric model assumptions. Text, image, and audio data present further challenges. To address such challenges, GDSC aims to explore transition systems for graph parsing of natural language and new fusion approaches for fully multimodal analysis. 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.