An Interdepartmental Computing Environment for Statistical Research

点击翻译按钮获取中文摘要

With support from the NSF SCREMS program, the Committee on Statistics at Arizona State University will purchase nodes in the university?s Fulton High Performance Computing Initiative that will guarantee access to high performance computing resources as well as dedicated server facilities. The equipment will be used for projects involving the development of parallel computing methods for a number of different statistical research problems including: i) computations for high-dimensional or sparse contingency tables, ii) constructing optimal designs for multilevel/multiresponse experiments with complex constraints, iii) constructing haplotypes from genotype data, iv) testing independence of parallel random number streams, v) developing resampling methods for inference in small area estimation with complex survey data, vi) implementing nonstationary multivariate spatio-temporal models using Monte Carlo Markov Chain methods, vii) developing new data mining methods and viii) estimation of non-standard (non-normal type) mixture models. The research enabled by the new equipment will provide a basis for future developments and research collaborations in many scientific areas. Statistical researchers at Arizona State University are housed in 7 different departments on 2 campuses. They are united through the framework of the university?s Committee on Statistics. The SCREMS-funded equipment will allow the Committee members to have a common computing environment for use in research and in training undergraduate and graduate students in statistics. Arizona State University statisticians are heavily involved in research that requires intensive computation. The new computing system will allow statistics students and faculty to have a common, state-of-the-art, framework for their research and to participate in the development of new computational methods for statistical design and analysis. Many of the Committee on Statistics faculty have close interdisciplinary connections, collaborating on problems in agriculture, biology, business, computer science, ecology, education, engineering, genetics, law, psychology, sociology and other areas. The computing equipment from this grant will allow the researchers to work on new computationally-intensive statistics methods for problems including: obtaining more accurate point and interval estimates for school-age poverty at the county level using data from the American Community Survey; studying how land management policies affect the spatial distribution of contaminants in soil; detecting diseases in MRI scans, attacks in computer networks, or deliberately released biological pathogens in the atmosphere; and constructing more efficient experimental designs for improving quality in industry.