Oct 23, 2012: Marco Pasveer: Importance sampling for probabilistic timed automata

October 23, 2012Importance sampling for probabilistic timed automata
Room: Zi 5126Marco Pasveer

In simulation we run a model several thousands of times and see whether or not certain event occurs. We can perform statistical tests on the results of the simulation to see how often the event occurs. When we are dealing with rare events, events with a very low probability of happening we need to run a lot of simulation runs before we can even detect the event.

This is where importance sampling comes in. With importance sampling we increase the probability that a certain rare event is triggered so we can reduce the amount of simulation runs needed to make an accurate guess. Afterwards the result is multiplied with a factor to account for the error we create when increasing the probability that a certain failure is triggered.

Importance sampling has already been applied to a wide range of models; however it has not been applied to probabilistic timed automata (PrTA) yet. I have been working on importance sampling for PrTA.