Efficiently Computing a Gradient by Structured Automatic Differentiation Can Save You Millions
Speaker:
Thomas Coleman, University of Waterloo
Date and Time:
Friday, June 3, 2016 - 4:00pm to 4:30pm
Location:
Fields Institute, Room 230
Abstract:
Recently there are been popular articles in the financial news claiming: : "Math Trick Saves Millions". The articles indicate how a "math trick" is reducing times to compute sensitivities of certain large portfolios of financial instruments by factors ranging from 50 to 50,000. In addition, this "math trick" is credited with savings millions in computing equipment and hardware. What is this "math trick"? It is in fact nothing more than careful design and application of (structured) reverse-mode automatic differentiation applied to expensive Monte Carlo portfolio computations. In this talk we full expose the "math trick".