CDS&E/Collaborative Research: The Integration of Data-Mining with Multiscale Engineering Computations

点击翻译按钮获取中文摘要

This collaborative research project aims at developing mathematical methods for the analysis/mining of large scale, high dimensional data that arise in engineering/physicochemical computations, and, more importantly, in exploiting these methods and algorithms to accelerate the computations themselves. This type of synergy between data mining and scientific computation has the potential to significantly enhance the way we extract knowledge from large scale modeling and simulation. The focus of the approach is the discovery a small number of key, intrinsic features of simulations characterized by very large numbers of degrees of freedom - and the exploitation of these key features to systematically design subsequent simulations. Another important feature of the work is the development of algorithms that "translate" between fine scale, detailed, and coarse scale, compact descriptions of the data, as well as algorithms for the fast incorporation of new data/information in previous computational frameworks. Deliverables of the effort will be the algorithms themselves, as well as documented illustrative examples of their application to large scale molecular and agent-based simulations.If successful, this research will create, document and make available a computational protocol for enhancing large scale scientific computations in the modeling of complex dynamical systems. Example applications include nanoscale self-assembly, such as micelle formation in materials computations, macromolecular foldling, large scale agent-based models of collective motion in ecology and cellular biology, as well as large scale, Partial Differential Equation simulations of engineering problems, like combustion. The results will be disseminated to allow their use by other researchers. The research will form the basis of cross-disciplinary education of graduate students in engineering and in mathematics, and of undergraduate research projects in these disciplines. It will also underpin the development of course materials in large scale data processing.