Scientific Computing Research Environments for the Mathematical Sciences (SCREMS)

宾夕法尼亚州立大学统计系提出了一个用于统计科学的计算环境的实施方案，支持五个学院的研究项目。这些项目对高速计算有着共同的要求，其应用包括天文学、图像分析、纵向数据分析、马尔可夫链蒙特卡罗（MCMC）算法、混合推理和环境科学。来自天文学、图像、生态学和地球科学的大量数据集需要大容量磁盘存储，相应的分析将大大受益于快速、并行计算资源，天文学方面的工作将加强和维持网络计算环境“VOSTAT”，这将使天文学家能够方便地对大量（兆兆字节规模）数据进行各种统计分析。数字图像的自动标注是一项极具挑战性的技术，具有重要的应用价值。将开发真实的时间算法，使用复杂的统计分类技术来自动分类图像。这种技术可以大大改善主要搜索引擎提供的图像搜索。纵向研究涉及随着时间的推移对实验单位进行重复观察的数据，需要专门的统计方法。将为这类研究调查新的统计方法。混合模型是对复杂数据建模的非常灵活的方法。一些理论和实践问题将在混合建模的背景下进行研究，特别是应用于海量数据集。MCMC算法是用于将现实模型拟合到复杂数据的计算工具。新的MCMC算法将在地理参考数据模型的背景下开发和研究。这些计算密集型算法也将用于与宾夕法尼亚州立大学地理学，生态学和气候学的合作。

The Department of Statistics at Pennsylvania State University proposes the implementation of a computing environment for the statistical sciences, supporting research projects of five faculty.These projects share a common requirement for high-speed computing with applications including astronomy, image analysis, longitudinal data analysis, Markov Chain Monte Carlo (MCMC) algorithms, mixture inference and environmental science. Massive data sets from astronomy, images and ecology and geoscience require high-capacity disk storage and the corresponding analyses benefit greatly from fast, parallelized computing resources.The work in astronomy will enhance and maintain the web computing environment `VOStat', which will allow astronomers to easily conduct a variety of statistical analyses on massive (terabyte scale) data. Automated annotation of digital pictures is a highly challenging technology with significant applications. Real time algorithms will be developed that use sophisticated statistical classification techniques to automatically classify images. This kind of technology can greatly improve upon image searches provided by major search engines. Longitudinal studies involve data with repeated observations on experimental units over time and require specialized statistical methods. New statistical approaches will be investigated for such studies. Mixture models are very flexible ways to model complex data. A number of theoretical and practical issues will be investigated in the context of mixture modeling, particularly as applied to massive data sets. MCMC algorithms are computational tools for fitting realistic models to complex data. New MCMC algorithms will be developed and studied in the context of models for data that are geographically referenced. These computationally intensive algorithms will also be used in collaborative work with Geography,Ecology and Climatology at Penn State.