TRIPODS: Algorithms for Data Science: Complexity, Scalability, and Robustness.

点击翻译按钮获取中文摘要

Award: CCF 1740551, Principal Investigator: Sham KakadeAlgorithmic tools underpin the ways in which modern data science methods glean insights from data, manipulate their environments, and estimate underlying statistical properties in the world. With increasing computational resources and an unprecedented growth of large datasets, there is an increased need for scalable and robust algorithmic tools which can provide insights into data in an automated manner, and thus, help to accelerate the pace of science and engineering. The modern challenges that a range of fields now face are no longer easily handled by ideas from a single discipline. A central goal of this project is to provide a common language and unifying methods for addressing contemporary data science challenges. At their core, each of the three disciplines of computer science, mathematics, and statistics has rich theories of complexity and robustness. These theories have influenced the design of the available tools that are used to address real world computational problems. Going forward, this project seeks new algorithms and design principles that unify ideas and provide a common language for addressing contemporary data science challenges. The PIs will draw from their expertise in computer science, mathematics, and statistics to aid in providing these unifying approaches. In parallel, aiming for a strong educational impact of the work, the aim is to train students an postdoctoral scholars to be well-versed in different areas underpinning data science and will incorporate appropriate theoretical ideas into a data science curriculum. The PIs will also organize events that help train students (including a hackathon and a bootcamp) and a research workshop that bring together researchers from the three disciplines for discussion and collaboration.In particular, the research objectives of this project are in unifying basic abstractions and techniques in order to yield not only further breakthroughs in all three fields, but also to impact societal and technological growth. The complexity and algorithmic questions this work seeks to address include: (i) how to unify various notions of complexity (which range from information theoretic to computational to black box oracle models), (ii) how to unify notions of robustness and adaptivity (e.g., how solutions and methods change as oracle models are corrupted by random or adversarial noise), (iii) how to address optimization challenges due to nonconvexity, and (iv) how to use these unified approaches to design more effective scalable tools, in theory and practice. These foundations will directly draw from the PIs close collaborations with various technological and scientific practitioners. Funds for the project come from CISE Computing and Communications Foundations, CISE Information Technology Research, MPS Division of Mathematical Sciences, and MPS Office of Multidisciplinary Activities.