Thursday 08 May 2014Adaptive Sticky Generalized Metropolis

Fabrizio Leisen - University of Kent

Plymouth University, Rolle 305 14:00-15:00

In the last decades, several adaptive Markov Chain Monte Carlo (MCMC) techniques have been developed using proposal densities that approximate adaptively the target distribution. Moreover, di fferent approaches have been proposed to update a proposal density within a MCMC algorithm. Here, we focus on the construction of sequence of proposals via interpolation procedures, using a subset of previous generated samples. The fi rst and the most famous methodology of this kind is surely the adaptive rejection Metropolis sampling (ARMS). A control test, checking the choice of interpolation points, must be added within the MCMC method to ensure the ergodicity of the chain and keep bounded the computational cost. We discuss diff erent possibilities, advantages and drawbacks of this strategy, within di fferent possible structures of MCMC methods.

Add to calendar

Add to calendar (.ics)