Innovative Computing

Multiscale modeling is the capability of representing different levels of phenomena, occurring at diverse spatial and temporal scales, within a unified conceptual framework. A typical example is the behavior of long molecules or nanosuspensions under the action of complex hydrodynamic flows. Our computational work tackles different phenomena such as biopolymer translocation through nanopores and hemodynamics in arteries with a novel computational scheme. The solution of such large-scale problems calls for the exploitation of powerful parallel architectures, such as the Blue Gene supercomputer, and sophisticated visualization tools to gain insight into the realm of multi-scale phenomena.

The development of new technologies that acquire large amounts of complex data is accelerating throughout medicine. Corresponding breakthroughs in accessible computation and algorithm development have made image analysis an indispensable tool for medical research and clinical practice. For example, image analysis enables the data acquired using diffusion tensor imaging (DTI) and functional magnetic resonance (fMRI) to reveal subject-specific structure and function of the brain. The emerging field of medical image computing requires strong, interdisciplinary teams of researchers, physicians, and engineers. Building such teams is a challenge but ultimately rewarding process. The Surgical Planning Lab at Brigham and Women's Hospital was founded in 1990 to enable research in image computation within the hospital context. Today, the SPL is the hub of the National Alliance for Medical Image Computing, a national effort with international impact across biomedicine and, increasingly, other fie

Modern science is increasingly faced with problems of ever greater complexity, straddling across the traditional disciplinary boundaries between physics, chemistry, material science and biology. Computational science is responding to this challenge with a steadfast development of innovative modeling techniques, designed in such a way as to offer an optimal handling of the information transfer procedures connecting the different scales/levels involved in the quantitative description of the aforementioned complex phenomena. This entails the seamless coupling between different mathematical representations of various physical phenomena at widely disparate scales, from continuum fields to probability distribution functions and atomistic trajectories, all the way down to many-body quantum wave functions. In this series of lectures, we shall provide an introduction to the basic ideas behind these triple-M (multiscale/multiphysics/multilevel) techniques, together with the illustration of a few

April 8, 2009 - Patrick Wolfe, Associate Professor of Electrical Engineering, Statistics and Information Sciences Laboratory, School of Engineering and Applied Science, Harvard University Modern science and engineering applications give rise to the vast quantities of high-dimensional data. This talk will provide a broad research perspective on the challenges and opportunities of drawing inferences from such data sets. For the large collections of sounds, images and networks acquired by modern sensing devices, traditional signal processing techniques singularly fail to scale, and new approaches are needed. Among the problems to be considered are forensic speech analysis, digital camera design and data reduction for large networks. Can we build practical solutions for these new contexts using the algorithms and tools of classical statistics?