CAREER: Coping with the Data Deluge - Algorithms, Human Aspects and Applications

在社会、科学、商业和军事应用中产生了大量的数据。然而，由于计算、带宽、功率限制以及注意力和隐私等人为限制，对数据的访问是有限的。因此，一个基本问题是以最小的成本获得最有用的信息。大多数现有技术都是没有保证的启发式技术，不能扩展或不能处理随时间变化的动态、分布式现象。此外，现有的算法通常忽略了信息收集任务中人的方面。本研究(1)为分布式、不确定、动态领域的优化信息收集开发有原则的算法；(2)研究优化信息收集的人类方面，将注意力、感知和隐私作为基本约束；(3)追求新颖的、现实世界的应用。为了实现这些目标，从根本上说，组合优化、概率推理和决策理论的新技术得到了发展。这些方法统一在子模函数优化的数学框架中。除了发展理论上有充分根据的、严谨的方法外，研究的一个重要部分是跨学科的评估，在社会计算、科学数据分析和可持续性随机优化方面的实际应用。这些优化信息收集的应用有可能为科学和社会提供新的系统和服务类别。通过与工业界（微软公司）和政府机构（美国地质勘探局和喷气推进实验室）的合作，以及通过综合教育计划、教程、共享数据和开源软件以及跨学科合作广泛传播结果，影响得以实现。

Massive volumes of data are now generated in social, scientific, business and military applications. However, access to the data is limited, due to computational, bandwidth, power limitations, as well as human constraints such as attention and privacy. A fundamental problem is thus to obtain most useful information at minimum cost. Most existing techniques are heuristics without guarantees, do not scale or cannot cope with dynamic, distributed phenomena that change over time. In addition, existing algorithms typically disregard human aspects of the information gathering task. This research (1) develops principled algorithms for optimized information gathering in distributed, uncertain, dynamic domains, (2) studies human aspects of optimized information gathering, considering attention, perception and privacy as fundamental constraints and (3) pursues novel, real-world applications. To accomplish these goals, fundamentally new techniques in combinatorial optimization, probabilistic reasoning and decision theory are developed. The approaches are unified in the mathematical framework of submodular function optimization. In addition to developing theoretically well-founded, rigorous approaches, an important part of the research is evaluation on interdisciplinary, real world applications in social computing, scientific data analysis and stochastic optimization in sustainability. These applications of optimized information gathering have the potential to enable new classes of systems and services for science and society. Impact is achieved through collaboration with partners from industry (Microsoft Corporation) and governmental facilities (USGS and JPL), as well as by broad dissemination of the results through an integrated education plan, tutorials, shared data and open-source software, and interdisciplinary collaborations.