Scientific Computing Research Environments in the Mathematical Sciences (SCREMS)

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

ABSTRACT PI: John A Rice Proposal: 0322751 The support from this grant allows the Department of Statistics at the University of California, Berkeley to purchase a Linux computer-server cluster of four PowerEdge dual processors, a 660 gigabyte StorEdge rack of disk drives, and a 20 cartridge rack mount tape library. This equipment is dedicated to the support of computationally intensive applied statistical research. In particular, the investigators and colleagues tackle problems in highway transportation, such as the calibration of micro models of driving behavior using video data of spatio-temporal segments of highways, taken from high above the roadway, a task requiring novel computer vision algorithms. They study questions in genomics and statistical genetics using gene expression microarray data, such as the identification of differentially expressed genes, the classification of tumors and other cell types, and the elucidation of gene regulatory mechanisms. They apply networks of temporal processes to neuroscience, and extend recent machine learning prediction methods and apply the results to cloud detection over ice and snow using multi-angle satellite image data. The research supported by this grant focuses specifically on projects in science sharing a common need for computationally intensive statistical methods to analyze large datasets. Such research requires computing facilities, which match the speed and storage capacity of the devices generating the datasets. The grant is to allow the Department to purchase appropriate processors and data storage devices to permit conduct of this research. The investigators and colleagues use this equipment to analyze video-recordings of traffic flow on interstate highways, in order to compare microscopic models of driver behavior with empirical data, the ultimate aim being better management of highway traffic. They use past data on forest fires to predict the likelihood of future fires as a function of location, past fire history, meteorological variables, burning indices, and other relevant variables. They develop statistical methods for the analysis of gene expression data collected to improve cancer diagnosis, to understand the genetic changes associated with disease susceptibility, and to contribute to basic research on cellular growth and development. They use multi-angle satellite image data to develop methods of predicting the presence of cloud over snow and ice, thereby permitting improved weather prediction and global warming monitoring. These examples constitute but a sample of the scientific research, which become feasible with the equipment purchased as the result of the grant support.