ATD: Understanding and Predicting User Mobility through Bayesian Models

该项目的目标是开发新的数学和统计工具，用于分析与人类流动模式相关的地理空间数据集。 重点是建模和预测的流动性，使用时空数据从一个中等到大量的演员，识别不寻常的行为模式或突然变化的既定模式（无论是在人口或个人层面），可以作为关键事件的主要指标的目标。 所提出的模型对社会学和人类学理论的评估，以及国防和自然安全的应用有影响。 第一个侧重于开发Mattern排斥点过程模型的变体，用于高分辨率地理参考数据，这些数据能够捕获现实生活中移动数据的众所周知的特征。 PI考虑了人口水平和个人水平的模型，这些模型具有高度的可解释性，并允许过程随时间的演变和系统中结构变化的识别。 第二个子项目侧重于（可能是隐藏的）关联数据的马尔可夫模型，其中只有粗略的地理空间信息可用（例如，可能知道个体在诸如无线接入点的固定点的范围内，但不知道其确切位置）。 同样，重点是识别个人和群体行为模式的变化，并对未来模式进行短期预测。 最后，第三个分项目侧重于将高分辨率和低分辨率地理空间数据的多个来源相结合以提高算法准确性的问题。 在所有情况下，重点是贝叶斯模型，它简化了不确定性量化，可以很容易地纳入正式的决策分析框架。

The goal of this project is to develop novel mathematical and statistical tools for the analysis of geospatial datasets associated with human mobility patterns. The focus is on modeling and prediction of mobility using spatio-temporal data from a moderate to large number of actors with the objective of identifying either unusual behavioral patterns or abrupt changes in well-established patterns (at either the population or individual level) that can serve as leading indicators of critical events. The proposed models have implications for the evaluation of sociological and anthropological theories, as well as for defense and natural security applications.The research revolves around three subprojects. The first one focuses on the development of variants of Mattern repulsive point process models for high-resolution geo-referenced data that are able to capture well-known features of real-life mobility data. The PI considers both population-level and individual-level models that are highly interpretable and allow for the evolution of the process over time and the identification of structural changes in the system. The second subproject focuses on (possibly hidden) Markov models for association data in which only coarse geo-spatial information is available (e.g., it might be known that the individual is within range of a fixed point, such as an wireless access point, but not their exact location). Again, the focus is on identifying changes in individual- and population-level behavioral patterns and making short-term predictions about future patterns. Finally, the third subproject focuses on the problem of combining multiple sources of high- and low-resolution geo-spatial data to improve the accuracy of algorithms. In all cases the emphasis is on Bayesian models, which simplify uncertainty quantification and can be easily incorporate into formal decision analytic frameworks.