Analysis of a Splitting Estimator for Rare Event Probabilities in Jackson Networks (with K. Leder and Y. Shi)

Summary:

Splitting and importance sampling are the most common techniques in rare event simulation. This paper shows that optimal splitting for Jackson networks runs in O(N^2*bottlenecks+1). This is the first refined analysis for splitting in the context of networks and it provides theoretical support for what has become a widely accepted fact within the community, namely, that splitting is relatively robust to the problem dimension.

Bibtex:

@Article{BlanLedShi09,

author = {Blanchet, J. and Leder, K. and Shi, Y.},

title = {Performance analysis of splitting algoritms for queueing networks},

journal = {To Appear in Stochastic Systems},

year = {2009},

volume = {},

pages = {}

}