Stochastic gradient systems: inference and applications

随机梯度系统：推理和应用

• 批准号: 0707157
• 负责人: David Brillinger
• 金额: $ 39万
• 依托单位: University of California-Berkeley
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2007
• 资助国家: 美国
• 起止时间: 2007-07-15 至 2011-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0707157&HistoricalAwards=false
• 关键词: Stochastic; gradient; systems; inference; applications

Data on the motion of objects has become common in many fields of science, engineering, and general human experience. Since the time of Newton mechanical motion has been described analytically by differential equations. Deterministic differential equations theory and application have developed into corresponding study of stochastic differential equations (SDEs). In a variety of practical situations it has been unclear how to select the drift function of an SDE. This research will investigate the use of a potential function to provide a formula for drift. In particular it will be assumed that the motion of interest is governed by a potential function. The drift is then the potential function's gradient. This structure is referred to as a stochastic gradient system. Being real-valued a potential function is easier to model than a vector-valued drift function. An estimated potential function may be used for simple description, summary, comparison, simulation, prediction, model appraisal, bootstrapping, and employed for estimating quantities of interest. The work will include study of models based on functional stochastic differential equations. This will allow the inclusion of time history in the description. Specific analytic problems to be studied include: unequally spaced times of observation, development of simulation methods to display variability, allowing model appraisal and making predictions. The theoretical structure investigated in the particle motion research will be applied to a variety of biological, ecological and other motion situations. In particular Hawaiian monk seal GPS data will be modeled and questions asked by the concerned marine biologists addressed. These data are important because the monk seal is America's most endangered marine mammal. Only about 1300 remain. The study of migration routes of northern elephant seal will also be modeled. This animal is a protected species under the Marine Mammal Act of 1972. Continuing, paths of animals in the Starkey Experimental Reserve in Oregon will be included in the work. These data have been collected to study the management of habitats shared by wild animals, cows and people. Lastly, work will continue on risk models concerning wildfires at the urban-wildfire interface will be modeled employing data from the San Diego County fires of 2003.

关于物体运动的数据在科学、工程和一般人类经验的许多领域中已经变得普遍。自牛顿时代以来，机械运动一直是用微分方程来解析地描述的。确定性微分方程的理论与应用已经发展到随机微分方程的研究。在各种实际情况下，如何选择陀螺仪的漂移功能一直不清楚。本研究将探讨使用一个潜在的功能，以提供一个公式的漂移。特别地，将假设感兴趣的运动由势函数支配。漂移就是势函数的梯度。这种结构被称为随机梯度系统。作为实值的势函数比向量值漂移函数更容易建模。估计的势函数可以用于简单描述、总结、比较、模拟、预测、模型评估、自举，并且用于估计感兴趣的量。这项工作将包括基于泛函随机微分方程的模型研究。这将允许在描述中包含时间历史。具体的分析问题进行研究，包括：不等间隔时间的观察，模拟方法的发展，以显示可变性，允许模型评估和预测。 在质点运动研究中研究的理论结构将被应用于各种生物、生态和其他运动情况。特别是夏威夷僧海豹GPS数据将被建模和有关海洋生物学家提出的问题得到解决。这些数据很重要，因为僧海豹是美国最濒危的海洋哺乳动物。只剩下1300人。还将对北方象海豹的迁徙路线进行建模研究。这种动物是1972年海洋哺乳动物法案中的受保护物种。继续，在俄勒冈州的斯塔基实验保护区的动物路径将包括在工作中。收集这些数据是为了研究野生动物、奶牛和人类共同栖息地的管理。最后，将继续进行关于城市野火界面野火风险模型的工作，将采用2003年圣地亚哥县火灾的数据进行建模。