|
Abstract
|
The development of multicore processors and graphic cards presents huge oppotunities for scientific computing. As fundamental techniques and tools, new numerical linear algebra algorithms based on loosely coupled parallelism(low synchronicity, low communication overhead) are strongly desirable. In this talk, we present such an algorithm for computation of leading paet singular value decomposition of very large, not necessarily sparse, matrices. Theoretical analysis and uses in several optimisation models are discussed.
|
