| Proj-1 |
Network reliability - Dr. Louis Petingi. The system under study is a communication network (e.g., wireless, optical, electronic networks) modeled by an undirected or a directed graph G = (V, E), where V is the set of nodes and E is the set of edges of G. Given a set of terminal nodes K of V (called participating nodes), and where each edge e fails with a known probability of failure qe, due for example to component’s wear-out or actions of intentional adversaries, the classical reliability measure, RK(G), gives the probability that the participating nodes will be able to communicate after deletion of the failed edges (communication links). In this project, students will be exposed to several theoretical reliability models and they will develop their own algorithms to measure the probability that the terminal nodes will be able to communicate between each other. Exact evaluation algorithms and heuristics will be developed to evaluate the reliability and as exact evaluation algorithms are computationally expensive (i.e., NP-hard), students will be also developing Monte Carlo techniques to estimate the reliability based on parallel processing libraries (MPI. OpenSHMEM). Simulation programs as NS3 will be run to validate the use of the proposed reliability measures. |
| Proj-2 |
Parallel collision search for cryptographic hash functions - Dr. Xiaowen Zhang. A cryptographic hash function (short for hash function) takes a much longer input message of arbitrary length and outputs a very shorter fixed-length bit-string, called hash. Since a large domain is mapped to a smaller range, collisions (pairs of inputs are mapped to the same output) are inevitable. However, as required for a hash function, it should be computationally infeasible to find any two distinct inputs that hash to the same value, i.e., collision resistant. Hash
functions are commonly used for data integrity in conjunction with digital signature. Parallel collision search for hash function is to find hash collisions in an efficient and effective way. |
| Proj-3 |
Routing policies of parallel/distributed particle filters - Dr. Feng Gu. Particle filters are important techniques to support data assimilation for large-scale systems. Parallel/distributed particle filters are used to improve the performance of particle filtering by distributing particles to multiple processing units. Two major steps are included in particle filters including sampling and resampling. The sampling procedure is easier to be parallelized due to no dependencies. However, resampling cannot be fully parallelized because it needs the global information of all the particles. After resampling, the selected particles need to be routed among different processors, which incurs a lot of communications, especially for large-scale systems. To improve the performance, we will investigate different routing policies to study their impacts on the performance of parallel/distributed particle filters . |
| Proj-4 |
GPU Acceleration for Digital Holographic Image Reconstruction and Processing - Dr. Shuqun Zhang. This project is to design, optimize and implement image processing algorithms for digital holography on GPU. The algorithms are image denoising, phase unwrapping, and phase shift and reference wave estimation. For the implementations, we will mainly focus on thread and memory management, and implementation strategies. We'll use CUNY's HPCC resources, in which Nvdia’s GPUs and CUDA platform will be used due to their capability and programmer friendliness. |
