Speaker

Pierre L'Ecuyer, Universite de Montreal, Canada;
DIRO, CIRRELT and GERAD

Abstract
Two novel algorithms for static network reliability estimation will be presented. The first one is based on a scheme that approximates zero-variance importance sampling. It provides bounded relative error, and even vanishing relative error in many cases, when the link reliabilities in the network converge to 1. The second algorithm uses a generalized splitting method on a Markov chain whose state extends the state of the network via auxiliary variables. It is typically the best performer when the network is large while the link reliabilities are fixed and not very close to 1. Numerical examples and comparisons will be provided.

Bio
Pierre L'Ecuyer currently holds a Canada Research Chair on "Stochastic Simulation and Optimization" at the University of Montreal. He received a PhD in operations research in 1983, from the University of Montreal. From 1983 to 1990, he was with the computer science department, at Laval University, Quebec. His research interests include Markov renewal decision processes, sensitivity analysis and optimization of discrete-event stochastic systems, and random number generation. His many awards include an INFORMS Simulation Society Outstanding Research Publication Award in 2009. He authored one of the 10 Landmark Papers that impacted the theory, education, and practice of simulation, selected among the nearly 10,000 papers of the first 40 years of the Winter Simulation Conference, 2007. Pierre is an INFORMS Fellow.