Rare Event Probability Estimation Using Information Projection

Anh Thu Phan Ho (1) , Wadih Sawaya (1) , Patrick Bas (1)

1 CRIStAL - Centre de Recherche en Informatique, Signal et Automatique de Lille - UMR 9189

This paper proposes an Importance Sampling (IS) simulation to estimate the rare event probabilities which satisfy a large deviation principle. The sampling distribution of the IS is constructed using the information projection. With this construction of the sampling distribution, the importance sampling simulation is showed to be asymptotically efficient. A specific example is presented to illustrate the IS technique.

Domaines: Traitement du signal et de l'image [eess.SP]

Anh Thu Phan Ho, Wadih Sawaya, Patrick Bas. Rare Event Probability Estimation Using Information Projection. International Symposium on Information Theory and Its Applications, Oct 2016, Monterey, CA, United States. ⟨hal-01359999⟩