A Loosely Coupled Distributed Singular Value Decomposition Algorithm for Large Dense Matrices

Title:  A Loosely Coupled Distributed Singular Value Decomposition Algorithm for Large Dense Matrices

 

SpeakerSheng Fang (University of Oxford)

 

Date16:00-17:00 pm, October 23, 2014

 

PlaceConference Room A304, Department of Mathematical Sciences

 

Abstract: The emergence of novel computational architectures, such as multi-core processors and graphic cards, makes it increasingly important to develop new highly parallel methods for big data computations that meet the loosely coupled computing requirements(low communication costs and low synchronicity) and harness the available computational power. Singular Value Decomposition(SVD) of large dense matrix plays an essential role in a variety of models and algorithms in data mining and machine learning. The calculations of SVD always form the computational bottleneck that dominates the computational costs in these models and algorithms. We will introduce a new methodology for the calculations of SVD of large and dense matrices that satisfies the loosely coupled computing requirements. We will start with a geometric motivation of our approach and a discussion of how it differs from other approaches. The convergence analysis is split into the derivation of bounds on the local error occurring at individual nodes, and bounds on the global error accumulation. Several variants of the algorithm will be compared, and numerical experiments and applications in matrix optimisation problems will also be discussed.

 

ContactLiping Zhang