Reconfigurable Acceleration of Multivariate Gaussian Generators

Speaker: Wayne Luk from Imperial College London Abstract:
This talk describes a hardware architecture for generating multivariate Gaussian distribution, which has been applied to financial risk modelling. The architecture contains two main elements: an efficient random number generator based on piecewise linear approximations, and a systolic matrix vector multiplier for introducing the required correlations. This approach has a raw generation rate of over 200 times that of a single 2.2GHz Opteron using an optimised BLAS package for linear álgebra computation. Practical performance is explored in a case study involving Delta-Gamma Value-at-Risk, where a single Virtex-4 xc4vsx55 at 400 MHz is 33 times faster than a quad Optaron 2.2GHz SMP. The FPGA solution also scales well, allowing larger portfolios to be simulated. Bio:
Wayne Luk is Professor of Computer Engineering at Imperial College London, and is currently Visiting Professor at Stanford University. His research interests include theory and practice of custom computing, reconfigurable architectures, and design automation.

Location: Hughes Aircraft Electrical Engineering Center (EEB) - -248

Audiences: Everyone Is Invited

Contact: Aimee Barnard