Collaborative Research: ATD: Statistical Detection of New Patterns and Potential Threats in Geospatial Sequences of Social and Political Events

该项目的重点是从统计上发现和解释社会和政治事件序列中的新趋势和地理空间模式变化。数以百万计的此类事件发生在地方、区域、国家和国际范围内。它们被记录在公开的领域，从新闻媒体到社交媒体、博客和推特等多种来源提供信息。事件包括公民骚乱、示威、犯罪、逮捕、人权活动、政治冲突、抗议、网络攻击、恐怖活动、新闻的发布、分析和讨论等，这些事件在随机时间发生，其地点、规模和后果都存在很大的不确定性。模式的突然变化、新分布或趋势的出现通常是由可能代表潜在威胁的新情况引起的。为了及时检测这种威胁，敏感而可靠的和计算上可行的统计算法，在地理空间序列中的变点检测将详细说明，其次是社会，经济和政治的统计上可检测的变化的解释。该项目将提供通用工具，以便对持续监测的大型数据库中发现的威胁性异常情况迅速作出反应，特别侧重于对国土安全构成潜在威胁的新模式。快速检测突发变化和意外出现的新趋势对于对潜在安全威胁做出迅速反应至关重要。为了在变点检测问题中处理高维的大数据集，结合联合收割机同时观测到的事件的地理空间序列，以及开发用于快速威胁检测的计算上可行的算法，利用了三种一般方法：（1）递归变点检测算法，其随着每个新的数据点更新，同时在每一步存储和处理最小所需信息;（2）辅助变点警报方案，其代表是用于早期检测潜在威胁的计算上廉价和快速的算法;（3）顺序规划的变点检测算法，其仅在特殊的中间时间点调用主检测方案;以及（4）借助于贝叶斯检测算法最大限度地使用先验信息。

The project focuses on the statistical detection and interpretation of new trends and geospatial pattern changes in sequences of social and political events. Millions of such events occur on a local, regional, national, and international scale. They are being recorded in publicly available domains, fed from a multitude of sources, from news media to social media, blogs, and tweets. Events include civic unrests, demonstrations, crimes, arrests, human rights activities, political conflicts, protests, cyber attacks, terrorist activities, publication of news, their analyses and discussions, and so on. Such events occur at random times, while their location, size, and consequences involve a lot of uncertainty. An abrupt change of pattern, appearance of a new distribution or trend is typically caused by a new circumstance that may represent a potential threat. For the prompt detection of such threats, sensitive yet reliable and computationally feasible statistical algorithms for change-point detection in geospatial sequences will be elaborated, followed by social, economic, and political interpretation of statistically detectable changes. The project will provide general tools for the prompt reaction to threatening anomalies identified in large continuously monitored databases, with a special focus on new patterns that represent potential threats to homeland security. Quick detection of sudden changes and unexpectedly appearing new trends is crucially important for the prompt reaction to potential security threats. To handle large data sets of high dimension in change-point detection problems, to combine simultaneously observed geospatial sequences of event data, and to develop computationally feasible algorithms for fast threat detection, three general approaches are exploited: (1) recursive change-point detection algorithms that are updated with each new data point while storing and processing minimum required information at each step; (2) auxiliary change-point warning schemes represented by computationally inexpensive and fast algorithms for the early detection of potential threats; (3) sequentially planned change-point detection algorithms that invoke the main detection scheme at the special interim time points only, and (4) maximum use of prior information by means of Bayesian detection algorithms.