HDR TRIPODS: UC Davis TETRAPODS Institute of Data Science

加州大学戴维斯分校的项目将建立加州大学戴维斯分校数据科学四足研究所(UCD4IDS)，该研究所将由来自四个系(计算机科学、电气与计算机工程、数学和统计学)的35名研究人员(4名PI和31名高级人员)组成，并将打破部门间的障碍，促进教职员工、博士后和研究生之间的跨学科研究合作。该项目将鼓励创新和强有力的研究，并为数据科学方面的研究生和博士后提供教育和指导。从事该项目的学生和博士后将被培养成下一代跨学科数据科学家：他们将获得一些重点领域的深入知识，同时拓宽他们在其他不同领域的视角。UCD4IDS将把四个主要系的教职员工的经验以及神经科学、医疗卫生科学和兽医等应用领域的经验引入UCD4IDS。UCD4IDS将组织：a)每周与数据科学有关的研讨会之后的圆桌讨论和分组会议；b)数据科学季度座谈会；c)每年三天的讲习班。该项目还将协调和开发加州大学戴维斯分校的各种课程，参与该项目的研究生将在四个系中的每个系至少学习一门课程。PI团队还将利用当地计划，通过将研究生、博士后和来自代表人数不足的群体的新教师与适当的导师相匹配，来招聘、支持和留住他们。为了传播研究和教育成果，PI团队计划：1)在网上提供座谈会和研讨会的演讲幻灯片、课堂讲稿和代码，这些将接触到我们当前和未来的合作者和公众；2)在目标会议上组织关于数据科学基础的小型研讨会和研讨会。UCD4IDS的研究将集中在三个广泛的主题上：1)面向生物和医学应用的机器学习基础；2)机器学习的优化理论和算法，包括大规模非平凡学习问题的数值解算器；以及3)图形和网络上的高维数据分析。即将开发的算法和软件工具将对解决不同领域的实际数据分析和机器学习问题产生积极影响，例如计算机科学(分析社交网络中的友谊关系)；电气工程(监测和控制传感器网络)；土木工程(监测道路网络上的交通流量)；特别是生物学和医学(分析在真实神经网络上测量的数据，检测疾病导致的大脑结构变化，为活生物细胞成像以分析其生长等)。这个项目的技术目标是：1)对高维数据的几何理解，这可能允许从代表某些感兴趣的现象的流形中高效(重新)采样，并对生物和医学应用中经常出现的细微但关键的差异进行分类；2)为非凸优化提供理论保证和高效的数值算法，这对机器学习至关重要；以及3)加深对单个实体(例如神经元)之间的局部相互作用如何导致全局协调和决策的理解。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 project at UC Davis will establish the UC Davis TETRAPODS Institute of Data Science (UCD4IDS), which will be composed of thirty-five researchers (four PIs and thirty-one senior personnel) coming from four departments (Computer Science, Electrical & Computer Engineering, Mathematics, and Statistics) and will break interdepartmental barriers and promote interdisciplinary research collaborations among faculty members, postdocs, and graduate students. The project will encourage innovative and robust research, and provide education and mentoring of graduate students and postdocs in data science. Students and postdocs engaged in this project will be trained to be the next generation of interdisciplinary data scientists: they will gain deep knowledge of some focused areas, and at the same time, broaden their perspectives in other diverse fields. The UCD4IDS will bring in the insights gained by the experience of the faculty members in the four primary departments as well as application fields such as neuroscience, medical and health sciences, and veterinary medicine. The UCD4IDS will organize: a) round-table discussions and breakout sessions after weekly seminars related to data science; b) quarterly colloquia on data science; and c) annual three-day workshops. The project will also coordinate and develop diverse courses at UC Davis, with graduate students involved in the project taking at least one course in each of the four departments. The PI team will also leverage local programs to recruit, support, and retain graduate students, postdocs, and new faculty members from underrepresented groups by matching them to appropriate mentors. For the dissemination of the research and educational results, the PI team plans to: 1) make colloquia and workshop talk slides, lecture notes, and codes available online, which will reach out to our current and future collaborators and the general public; and 2) organize mini-symposia and workshops on foundations of data science at targeted conferences.Research at the UCD4IDS will focus on three broad themes: 1) Fundamentals of machine learning directed toward biological and medical applications; 2) Optimization theory and algorithms for machine learning including numerical solvers for large-scale nontrivial learning problems; and 3) High-dimensional data analysis on graphs and networks. The algorithms and software tools to be developed will make a positive impact in solving practical data-analysis and machine-learning problems in diverse fields, e.g., computer science (analyzing friendship relations in social networks); electrical engineering (monitoring and controlling sensor networks); civil engineering (monitoring traffic flow on a road network); and in particular, biology and medicine (analyzing data measured on real neural networks, detecting changes in the brain structures due to diseases, imaging live biological cells for analyzing their growth, etc.). The technical goals of this project are: 1) geometric understanding of high-dimensional data, which may allow efficient (re)sampling from manifolds representing certain phenomena of interest and classifying subtle yet critical differences that often appear in biological and medical applications; 2) providing theoretical guarantees and efficient numerical algorithms for non-convex optimization, which is crucial to machine learning; and 3) deepening understanding of how local interactions between individual entities (e.g., neurons) lead to global coordination and decision making. 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.