MCMC-driven Adaptive Multiple Importance Sampling

Title:

 

MCMC-driven Adaptive Multiple Importance Sampling

 

Abstract:

 

Monte Carlo (MC) methods are widely used for statistical inference and stochastic optimization. A well-known class of MC methods is composed of importance sampling (IS) and its adaptive extensions (such as adaptive multiple IS and population MC). In this work, we introduce an iterated batch importance sampler using a population of proposal densities, which are adapted according to an MCMC technique over the population of location parameters. Compared with a traditional multiple IS scheme with the same number of samples, the performance is substantially improved at the expense of a moderate increase in the computational cost due to the additional MCMC steps. Furthermore, the dependence on the choice of the cloud of proposals is sensibly reduced, and the proposal density in the MCMC method can be adapted in order to optimize the performance.

 

 

About the presenter:

 

Luca Martino obtained his PhD in Statistical Signal Processing in the Universidad Carlos III de Madrid, Spain, in 2011. Currently, Luca is a postdoctoral researcher within COIN