Rare event simulation : a point process interpretation with application in probability and quantile estimation Walter, Clément; Defaux, Gilles
This paper addresses the issue of estimating extreme probability and quantile on the output of complex computer codes. We introduce a new approach to solve this problem in term of a random walk in the output space, leading to two main results: (1) the number of samples required to get a realisation of a random variable in a given domain of probability measure p is drastically reduced, following a Poisson law with parameter log1/p; and (2) we get parallel algorithms for estimating probabilities and quantiles and especially the optimal parallel Multilevel Splitting algorithm where there is indeed no subset to define anymore.
Item Citations and Data
Attribution-NonCommercial-NoDerivs 2.5 Canada