Optical projection tomography (OPT) is a novel approach for three-dimensional (3-D) imaging in biology using optics where the specimens could be of size between 1 mm to 1 cm. The image reconstruction uses simple back projection and has been largely debated about the quantification provided by these techniques being inadequate to do biological studies. This project primarily involves development novel computational methods to improve the OPT, including iterative reconstruction techniques based on compressive sensing.

Medical images are blurred due to organ motion and nature of radiation interaction with tissue. In both cases, the convolution point spread function (PSF) for the blur can easily be estimated. The deblurring of these images using techniques like Truncated Singular Value Decomoposition (TSVD) will be attempted using both simulation and real-time data. The improvement by deblurring of medical images in terms of contrast and resolution will be quantified.

The aim of this work is to develop an efficient and robust parallel (MPI) algorithm for partitioning hierarchy of computational mesh using Metis/ParMetis package for parallel Multigrid solvers. This work also involves the implementation of conforming, nonconforming and discontinuous finite element spaces on the hierarchy of partitioned mesh. The efficiency and the robustness of the algorithm will be tested on complex geometries such as human nasal cavity, crystallizers, etc.

The aim of this work is to develop parallel iterative solvers for large sparse linear systems arising from the finite element discretization of elliptic partial differential equations. The developed parallel iterative solvers will be used in the context of geometric multigrid method. The study on the efficiency in terms of computational time, communication time will be performed, in particular for the incompressible Navier-Stokes system.

Liquid droplet spreading or recoiling over a solid surface induces a dynamic contact angle with respect to the contact line velocity. The slip coefficient arising form the slip boundary condition on the liquid-solid interface plays a significant role in this relation. Here, the influence of slip coefficient on the dynamic contact angle will be studied numerically. Simulations for a 3D-axismmetric liquid droplet which impinge and deforms on a horizontal solid surface will be considered for this study. The finite element method with the Arbitrary Lagrangian-Eulerian (ALE) approach will be used in the the simulation.

Typical computer vision application such as object detection, tracking and eventdetection often exhibit poor performance for analyzing crowd videos. This project aims to develop detection and tracking algorithms that are suitable for crowd flow analysis and recognize the anomaly behavior from normal behavior in real-time.

Every day millions of hours of video are captured around the world by CCTV cameras, webcams, and traffic-cams. Most of these videos, especially surveillance videos, are just archived and not watched or analyzed by anyone. Most of these videos are non eventful and monotonous. This project aims to represent these videos compactly, which features only activities of interest while preserving the general dynamics of the original video.

Increasing the resolution of the images without artifacts is useful for various applications. This project aims to develop sparse representation based algorithms for image and video superresolution. Further the algorithm will be extended for image denoising.

This project is on developing novel algorithms/data structures for large graphs on GPUs and their performance tuning.

No. of Students : 1

Optical processes that involve absorption and emission at different energies (Photo-luminescence, Fluorescence) are non-linear typically. Using commercial Finite-Difference-Time-Domain (FDTD) tools we aim to model such processes in composite materials. This project would involve learning to use the FDTD tools using worked examples and developing them into non-linear models. Requires familiarity with basic optical science.

Topological invariants often represent the characteristics of geometric structures. Computing these invariants is important in many scientific domains. Largertopologies result in larger computation times for the invariants. In this project, parallel algorithms will be developed to speed up the computations of the invariants. The algorithms will be evaluated for different data domains and ondifferent parallel platforms including BlueGene/L and GPGPUs.

This project will analyze the job workload traces of parallel systems, identify and predict applications and system performance, and develop job scheduling strategies based on the predictions and analyses.

Multi-physics parallel applications are composed of multiple component applications, each of which is a parallel application, and interacts with other component applications. This project will consider developing programming abstractions and runtime execution strategies for these applications.

This project will develop framework including user interfaces and runtime systems that perform automatic optimizations of applications on GPUs, including selecting granularity of a gpu thread, otpimizing data layout on GPUs, and adaptive asynchronous operations between CPU and GPU computations. The framework will be used for optimizations of applications including linear algebra and adaptive applications.

This project will develop efficient algorithms for implementing collective communications of the latest MPI 3 standard, and will also develop algorithms for collective communications tunes for torus networks.

Climate Modeling has significant computation intensive calculations including (but not limited to ) radiation calculations and parameterization of cloud-effects. This project will explore speeding up these intensive calculations on GPUs by developing algorithms and strategies.

Ensemble Kalman Filtering (EnKF) is a technique used for developing initial conditions for numerical weather/climate prediction models. EnKF involves exploration of the input parameter space involving multiple realizations with the predictive model/filter. The project will explore using computational grids for EnKF and development of relevant algorithms/middleware for exploiting grids.