Title: Accelerating Matrix Decomposition Problems By Using Reconfigurable Computing Platform
Speaker: Xinying Wang, ECpE Graduate Student
Advisor: Joseph Zambreno, Associate Professor
Abstract: Matrix decomposition such as singular value decomposition, eigenvalue decomposition, QR decomposition and sparse LU decomposition are widely used in many scientific and engineering applications, such as signal processing, text mining and wireless communications. However, matrix decomposition has been considered as a computationally expensive process, and their sequential implementations fail to meet the requirements of many time-sensitive applications. While the performance of software implementations is restricted by their inherent computational complexity, the emergence of FPGAs provides a flexible and low-cost opportunity to pursue parallel high-performance matrix decomposition designs. In this presentation, I will introduce our FPGA-based architectures for singular value decomposition, eigenvalue decomposition, QR decomposition, and sparse LU decomposition. Meanwhile, experimental results using an FPGA-based hybrid acceleration system will also be presented to demonstrate the efficiency of our novel architectures. Moreover, the potential applications of those designs will be analyzed.