Efficient sampling of rare events by splitting
Seminar Room 1, Newton Institute
AbstractStandard (or crude) Monte Carlo (MC) simulation is known to be inefficient for simulating rare events. For events with low probability, the squared relative error on estimates obtained from straightforward MC simulation is inversely proportional to the number of samples, so that an excessively large number of samples may be required to reach a desired accuracy for the estimation of rare event probabilities.
To improve the efficiency of MC sampling for rare events, various techniques have been developed in the past, for applications in e.g. communication networks and reliability analysis. Such techniques can be of interest for studying extremes in geophysical models. I will discuss a technique called multilevel splitting, in which model sample paths are split into multiple copies each time they cross thresholds (or levels) that lead closer to the rare event set.
If it doesn't, something may have gone wrong with our embedded player.
We'll get it fixed as soon as possible.