Importance Sampling for Estimation of Small Probabilities.
NAVAL UNDERWATER SYSTEMS CENTER NEW LONDON CT NEW LONDON LAB
Pagination or Media Count:
The use of importance sampling for estimation of small probabilities is illustrated by means of a signal-detection example. In particular, false alarm probabilities in the ten to the minus 7th power range are accurately estimated by means of only 1000 independent trials. The fundamental variance-reducing capability of importance sampling is explored and used to significantly improve the performance of the probability estimate for the detection example considered. Guidelines for choosing good data-generation algorithms are presented. Author
- Statistics and Probability