Decision Making in the Face of Uncertainty: Comparative Probability Metrics and Network Surveillance

当今社会充斥着数据，这场数据革命正在改变决策的方式；在所有自然科学和工程(NSE)领域，面对不确定性做出决策，并使用数据为这些决策提供信息，正变得司空见惯。在这里，我提出了一个研究计划，其主要目标是开发工具和方法，以促进此类决策过程，并在两个特定背景下为NSE研究人员提供清晰和信心。 首先，我们考虑要评估两个或两个以上特征的相似性或优越性的比较。传统上，这样的比较是通过假设检验进行的，对于这些假设，零假设假定相等。然而，在实践中，通常更自然的是假设这些特征是不同的，直到有足够的证据证明它们是平等的。重要的是，在这种情况下，精确对等并不总是必要的，相反，实际对等就足够了。这两个考虑因素都被我称为一致概率和其他量身定做的比较概率度量(CPM)优雅地处理了。这些衡量标准和这种思维方式构成了一种创新和优雅的手段，在考虑实际重要性的同时，正式评估和量化相似性和优越性。在这里，我提出了几种新的基于CPM的策略，用于各种应用领域，包括比较测量设备、寿命分布和在不同实验条件下的结果。我希望这个研究项目将导致决策策略的范式转变，并取代假设检验作为进行此类比较的标准方法。 其次，我们考虑了网络监控领域，它涉及到识别社交网络中的异常活动。随着近年来社交网络的普及，网络监控的重要性也与日俱增。我的工作一直致力于开发实时监测方法，试图确定网络中的互动何时以及如何发生重大变化。这些技术已经成功地识别了垃圾邮件发送者、恐怖分子和社交网络中的欺诈活动。这一研究领域方兴未艾，有许多开放的研究问题和监测策略需要探索。在这项提议中，我强调了我的计划，即制定一种统一的方法，快速准确地检测异常情况，同时考虑到社区结构、个人互动倾向、互动方向以及可能有用的背景信息。这种方法将推动当前最先进的技术，并极大地提高我们识别社交网络中邪恶活动的能力。 这项研究计划将包括研究生和本科生，他们都将进行及时和有影响力的研究，并获得非常受欢迎的技能。

Society today is inundated with data and this data revolution is changing the way decisions are being made; in all natural science and engineering (NSE) fields making decisions in the face of uncertainty and using data to inform these decisions is becoming commonplace. Here I propose a research program whose main objective is to develop tools and methodologies that facilitate such decision-making processes and that provide NSE researchers with clarity and confidence in two specific contexts. First, we consider comparisons in which the similarity or superiority of two or more characteristics is to be evaluated. Traditionally, such comparisons are made via hypothesis tests for which the null hypotheses assume equality. However, in practice it is often more natural to assume the characteristics are different until there is sufficient evidence to justify considering them equal. Importantly, in such contexts exact equivalence is not always necessary and instead practical equivalence is sufficient. Both of these considerations are gracefully handled by what I call the probability of agreement and other tailored comparative probability metrics (CPMs). These metrics and this way of thinking constitute an innovative and elegant means of formally evaluating and quantifying similarity and superiority while accounting for practical importance. Here I propose several novel CPM-based strategies for use in a variety of application areas including the comparison measurement devices, lifetime distributions, and outcomes in different experimental conditions. I expect that this research program will lead to a paradigm shift in decision-making strategies and to the replacement of hypothesis testing as the standard approach for making such comparisons. Second, we consider the field of network surveillance which is concerned with identifying anomalous activity in social networks. As the prevalence of social networks has increased in recent years, so also has the importance of network surveillance. My work has been devoted to the development of real-time monitoring methodologies that seek to identify when and how interactions within a network significantly change. Such techniques have been successful in identifying spammers, terrorists and fraudulent activity in social networks. This research area is burgeoning with many open research problems and monitoring strategies to explore. In this proposal I highlight my plans to develop a unified methodology that quickly and accurately detects anomalies while accounting for community structure, individual propensities for interaction, the direction of interactions as well as contextual information that might be useful. Such a methodology will advance the current state-of-the-art and greatly improve our ability to identify nefarious activity in social networks. This research program will involve both graduate and undergraduate students who will all conduct timely and impactful research and acquire highly sought-after skills.