Computational methodologies for state-space models: A Big-Data challenge

状态空间模型的计算方法：大数据挑战

• 批准号: 2119410
• 金额: --
• 依托单位: University of Kent
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2018
• 资助国家: 英国
• 起止时间: 2018 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2119410
• 关键词: Computational; methodologies; state; space; models

The ready availability of data requires improved methods for model fitting. Within this project, we will explore Bayesian approaches to handle the large class of state-space models, with a particular focus on model design, to capture the information contained in large amounts of data. High dimensionality and large sample sizes pose serious challenges in model selection and inference. In fact, model components such as the likelihood itself and the state transformation density, are usually intractable under the above conditions and regular approaches based on Bayes factors, for example, cannot be implemented. In this project, new approaches to deal with Big-Data problems within the Bayesian context will be explored and developed with the aim of overcoming modelling and current computational limitations. Besides methodological investigations, supported by simulated data, the project will also involve practical applications, in particular to the areas of finance and ecology.

数据的现成可用性要求改进模型拟合方法。在这个项目中，我们将探索贝叶斯方法来处理大类别的状态空间模型，特别关注模型设计，以捕获包含在大量数据中的信息。高维数和大样本量对模型选择和推理提出了严峻的挑战。实际上，在上述条件下，模型成分（如似然本身和状态转换密度）通常是难以处理的，例如基于贝叶斯因子的常规方法无法实现。在这个项目中，将探索和发展贝叶斯背景下处理大数据问题的新方法，以克服建模和当前的计算限制。除了在模拟数据的支持下进行方法学调查外，该项目还将涉及实际应用，特别是在金融和生态领域。