Motivated by real problems in the IT industry like data center/supercomputer thermal management and simulations of business operations, the investigators aim to develop a set of novel statistical methods for the experimental planning, modeling, and optimization of modern IT systems. Specifically, they propose new modeling techniques for IT systems with high-dimensional responses, with spatial and temporal responses, or with a large number of configuration variables, and for the experimental planning and optimization of business operation simulations. The investigators bring together expertise in computer modeling and experiments, latent variable and structural equation models, modern optimization, and design of experiments. Their research will bridge the gap between statistical practice in academe and industry. They will use challenging real-world problems at IBM to motivate and develop new methodologies, and in turn test them on real data for improvement. Its intellectual merit lies in the development of a general methodology for designing, modeling, and optimizing IT or other systems which exhibit similar traits to those described above. It can lead to major advances in multivariate statistical modeling, stochastic process modeling, statistics-aided stochastic optimization, optimization-aided design search, and space-filling designs. It is conceivable that the methodology can be incorporated into publicly released software, thus directly benefiting practitioners and researchers.
Many new IT systems like supercomputers, data centers, and storage systems have been invented to accommodate different business and personal needs. A recent trend in the IT industry is the integration of the traditional business in hardware with enterprise IT services like web hosting, data storage, and high-performance computing. The proposed work is expected to have broad-based impacts on the designing, monitoring and management of complex systems in general. It can be applied to a variety of problems like power systems and aircraft engine combustors. It can also benefit large-scale computer modeling for climate change, reaction to natural crisis, and the spread of pandemic diseases like bird flu and SARS. Specifically, the developed methods can improve the thermal management and cooling efficiency of electronic systems, which is a pressing issue faced by many US industries today. The team is committed to education and dissemination of the new methodology to make a long-term impact. They plan to infuse research outcomes into coursework. The proposed education plan will result in rigorous training of a diverse group of students with solid background in statistical methods, decision-making under uncertainty, and computational modeling, who will meet a critical need of IT and other high-tech industries. They are committed to creating a diverse environment in their research groups in terms of race, gender and national origin.
