Frontiers in Bayesian methods for computationally intractable models

计算棘手模型的贝叶斯方法前沿

• 批准号: DP160102544
• 负责人: Prof Scott Sisson
• 金额: $ 28.33万
• 依托单位: The University of New South Wales
• 依托单位国家: 澳大利亚
• 项目类别: Discovery Projects
• 财政年份: 2016
• 资助国家: 澳大利亚
• 起止时间: 2016-01-01 至 2018-12-31
• 项目状态: 已结题
• 来源: https://dataportal.arc.gov.au/RGS/Web/Grants/DP160102544
• 关键词: Frontiers; Bayesian; methods; computationally; intractable

This project aims to develop new statistical methods for the analysis of computationally intractable models. Direct statistical analyses are often impossible when models are highly complex or too large, or when the dataset is too big. Using simpler models means that the wrong questions are being answered. Using less data is wasteful of information. This project aims to develop new methods for Bayesian statistical inference when standard methods are intractable. By allowing otherwise unavailable analyses, such techniques can enable and accelerate across-the-board research advances. Key expected innovations are new bases for sampling from computationally intractable distributions and novel distributed computing algorithms which could be applied to problems in health economics to infectious disease modelling, from climate extremes to road traffic modelling.

这个项目的目的是开发新的统计方法来分析难以计算的模型。当模型非常复杂或太大时，或者当数据集太大时，直接统计分析通常是不可能的。使用更简单的模型意味着回答了错误的问题。使用较少的数据是信息的浪费。这个项目的目的是开发新的贝叶斯统计推断方法，当标准方法难以处理时。通过允许其他方面无法获得的分析，这种技术可以促进和加速全面的研究进展。关键的预期创新是从难以计算的分布中进行采样的新基础，以及新的分布式计算算法，这些算法可以应用于卫生经济学中的问题，到传染病建模，从极端气候到道路交通建模。