Spatio-Temporal Data Mining and Artificial Intelligence for Computer Animation

在智能手机和平板电脑等移动设备上运行的应用程序(App)正在生成海量的时空数据数据库。在时空数据中，诸如由移动设备的全球定位系统(GPS)给出的物理位置以及诸如用户在该位置的时间是重要的。例如，如果这款应用分享了蚊子密度的信息，用户报告遭遇蚊子肆虐的地点和时间就很重要。在我的研究中，我将开发新的技术来对时空数据进行数据挖掘，然后将这些想法在软件中实现。数据挖掘是指在一台计算机上运行的自动化过程，通常是一台与应用程序通信的较大的中央计算机。它接受一些数据作为输入，在数据中找到模式，然后在屏幕上显示结果或将结果传递给其他软件。虽然可能会考虑许多类型的模式，但我将集中讨论两种类型，称为高效用项集和有趣的长方体。在这个总结中，我将只描述第一种类型。高效用项集是一些一起报告的项的组合(例如，购买中的食品杂货项的组合或在城市交叉点观察到的特征)，这对在数据中寻找模式的人非常有用。因此，想要执行数据挖掘的人决定哪些属性将使组合非常有用，然后数据挖掘软件在数据库中搜索以找到具有这些属性的所有组合。在我的研究中，我将特别寻找在某些地方和某些时间发生的组合。一个挑战将是确定一个地方有多大(它的大小因不同的问题而有所不同)以及应该考虑的持续时间(它也会有所不同)。你可以想象一家连锁便利店使用我们的研究结果来决定应该向哪些移动用户提供哪些特别优惠(现在就一起购买这两种商品！)以及在什么地点和时间。如果成功，这项研究将带来新的软件产品，这些产品将立即引起零售商和广告商的兴趣，并可能对商业上不那么商业化的众包应用程序有用。尽管它没有密切相关，但我也将继续进行好奇心驱动的研究，将人工智能技术应用于模拟大量动物。例如，生物学家对座头鲸如何能够共同努力将鱼赶到一个被气泡窗帘包围的紧凑球体中，然后突然所有的鱼从鱼的下面涌出并吃掉它们提出了许多问题。由于观看这些事件的发生既困难又昂贵，我们将创建在计算机上运行的事件的动画模拟。我们将编码生物学家关于鲸鱼可能正在做的事情的想法，并进行模拟，看看它们是否有预期的效果。由于生物学家将能够看到动画鲸鱼正在做动作，他们将能够判断他们的想法是否正确，以及我们的模拟是否正确。需要模拟的数十万条鱼对我们的编程将是一个挑战，我们希望设计一些新的技术来使模拟足够有效，以它们的自然速度运行。我们的研究结果将引起生物学家的兴趣，特别是那些对座头鲸感兴趣的人，也包括那些对其他类型的动物感兴趣的人，比如蝙蝠，这些动物形成了一个大群体。这些技术在新的电脑游戏和动画电影中也可能有用。

Applications (apps) running on mobile devices, such as smart phones and tablet computers, are generating massive databases of spatio-temporal data. In spatio-temporal data, a physical location, such as one given by the Global Position System (GPS) of the mobile device, and a time, such as when the user was at that location, are important. For example, if the app shares information about mosquito density, the place and time where a user reports encountering a heavy mosquito infestation are important. In my research, I will develop new techniques for performing data mining on spatio-temporal data and then implement these ideas in software. Data mining refers to an automated process that runs on a computer, typically a larger central computer that the app communicates with. It takes some data as input, finds patterns in the data, and then either displays the results on the screen or passes them to some other software. Although many kinds of patterns might be considered, I will concentrate on two types, called high utility itemsets and interesting cuboids. In this summary, I will describe only the first type.A high utility itemset is a combination of some items that were reported together (such as a combination of grocery items in a purchase or features observed at a city intersection) that is highly useful to the person looking for patterns in the data. Thus, the person who wants to perform data mining decides what properties would make combinations highly useful and then the data mining software searches through the database to find all the combinations that have those properties. In my research, I will look in particular for combinations that happen at some places and some times. A challenge will be to identify how large of a place (it will vary in size for different problems) and how long a duration (it will also vary) should be considered. One can imagine a chain of convenience store using the results of our research to decide which mobile users should be offered which special offers (buy these two items together right now!) and at what places and times. If successful, this line of research will lead to new software products that will be of immediate interest to retailers and advertisers and may also be useful in less commercial crowd-sourced apps.Although it is not closely related, I will also pursue curiosity-driven research on applying Artificial Intelligence techniques to simulating large numbers of animals. For example, biologists have many questions about how humpback whales are able to work together to herd fish into a compact ball surrounded by a sort of curtain of bubbles and then suddenly all surge up together from under the fish and eat them. Since it is difficult and expensive to see these events occurring, we will create an animated simulation of the events running on the computer. We will encode ideas from the biologists about what the whales may be doing and run the simulation to see if they have the expected effect. Since the biologists will be able to see the animated whales doing the actions, they will be able to judge whether their ideas were correct and whether our simulation is correct. The hundreds of thousands of fish that need to be simulated will be a challenge for our programming and we expect to devise some new techniques to allow the simulations to be efficient enough to run at their natural speed. The results of our research will be of interest to biologists, especially those interested in humpback whales but also those interested in other types of animals, such as bats, that form large groups. The techniques may also be useful in new computer games and animated films.