I am a computer science Ph.D. student at the University of Michigan. The primary focus of my work has been pushing the envelope of computing by using parallel computers to solve problems with extensive space and time requirements.

One area in which I am working is in the calculation of optimal adaptive statistical designs. These designs have significant ethical and cost advantages over standard statistical designs. Despite these advantages, however, adaptive designs are rarely used because they are extremely computationally expensive to optimize and analyze. For example, the algorithm for optimizing one design we are interested in has the formidable growth rate of O(n^6) in space and time. Despite this we have been able to solve this problem for inputs of nontrivial size, a feat long considered infeasible by the statistics community. In some cases, we have solved problems thousands of times more difficult than previously considered possible, using up to 32 processors with a combined memory of 32 GB. Our eventual goal is to remove the computational impediments to using adaptive statistical designs, so that they can be more widely used. I hope that some of the tools in your workshop will bring us closer to our goal. This interdisciplinary work involves statisticians and computer scientists, and is funded by NSF.

Another area of research in which I am involved is a project that aims to develop a predictive space weather model. This very large interdisciplinary project is funded by NASA and through the KDI program of NSF. It is headed by Tamas Gombosi, a space scientist here at the University of Michigan, and includes experts in fluid dynamics, numerical analysis, climate modeling, upper atmospheric models, and computer scientists. Partners include the National Center for Atmospheric Research (NCAR), Rice University, Sterling Software, Texas Southern U., and Northern Michigan University.

Geomagnetic storms are large scale disruptions in the earth's magnetic field. They can often destroy satellites or knock out power distribution systems. These storms are caused by events, such as solar flares, on the sun. Our goal is to use data gathered from solar observatories to extrapolate how events on the sun will effect the heliosphere and in turn how the heliosphere will effect the earth's magnetic field and atmosphere. Our hope is to be able to predict geomagnetic storms so that protective steps can be taken before they occur.

The initial software to perform this prediction will consist of three separate adaptive mesh systems between which a complex web of data and control will flow. On its own each system will be quite complex, and when combined they will be much more so. It is hard to quantify the exact amount of computer power that will be needed for the space weather prediction software, but given that we need to do accurate predictions faster than real time, it will be considerable, on the order of a few hundred processors.

In conjunction with my advisor, I am responsible for specifying and developing the parallel data structures for this project. The tools presented at your workshop may help us to simplify the construction of our software and help us to reach our goal of accurate space weather prediction more quickly.

In addition to discovering the uses of your existing tools, attending the workshop would provide me with a valuable example of how such tools are constructed. This knowledge will allow me to develop similar tools from my own research.