CAREER: Algorithmic Foundations for Social Data

在过去的几年里，大规模数据带来了大量的机会和希望。尽管在十多年的时间里已经收集和分析了大量的数据集，但令人兴奋的主要是由于相对较新的社会数据的可用性：人类互动的大量数字记录。这为人类集体行为提供了一个独特的全系统视角，这构成了根本性的挑战和机遇。尽管近年来取得了巨大的进步，但迄今为止，很少有算法框架专门用于分析社会数据集。这个项目的目标是开发能够分析大规模社会数据的框架。该项目寻求具有丰富问题的新颖模型，提出关于计算的深刻问题，并可能对社会学和数据科学产生长期影响。从技术角度来看，该项目的目标是开发具有强大理论保证的适当算法机制，并将其转化为实践中的结果。该项目包括三个主要的研究方向。研究的第一线是寻求发展一种理论，在给定我们现在采取的行动的后果分布的情况下，优化未来的事件。第二条研究线考虑了社会数据的可学习性和可扩展性，以及其优化解释。研究的第三条线考虑设计针对噪声数据的鲁棒优化算法。该方法包括在真实数据集上进行实验，以开发具有强大理论保证的适当算法机制，并将其转化为实践中的结果。本科和研究生课程都将受益于这一跨学科领域的课程发展。

In the past several years, there has been a great deal of exposure to the opportunities and promise that lie in large-scale data. Although massive data sets have been collected and analyzed in well over a decade, the excitement is largely due to the relatively recent availability of social data: massive digital records of human interactions. This provides a unique system-wide perspective of collective human behavior which poses fundamental challenges and opportunities. Despite the tremendous progress made in recent years, very few algorithmic frameworks to-date have been purposefully developed for analyzing social data sets.  The goal of this project is to develop frameworks that enable analysis of large-scale social data.  This project seeks novel models that are rich in problems, raise deep questions about computation, and can lead to long-lasting impact on sociology and data science.  From a technical perspective, the goal of the project is to develop appropriate algorithmic machinery with strong theoretical guarantees that translate to results in practice.  The project consists of three main lines of research.  The first line of research seeks to develop a theory to optimize events in the future given a distribution on the consequences of actions we take in the present.  The second line of research considers learnability and scalability of social data, and its interpretation for optimization.  The third line of research considers design of robust optimization algorithms for noisy data.  The methodology includes experimentation on real data sets to develop appropriate algorithmic machinery with strong theoretical guarantees that translate to results in practice. Both undergraduate and graduate curriculum will benefit from the development of courses in this interdisciplinary area.