Date & Time: Wednesday, August 06, 2014, 16:00-17:00.
Venue: Ramanujan Hall
Speaker: Krishna B. Athreya, Iowa State University
Title: Regenerative sequence Monte Carlo methods for improper target distributions
Abstract: If $\pi$ is a probability distribution on some set $S$ and $f$ is a real valued function with $E|f(X)|<\infty$ where $X$ is a random variable with distribution $\pi$ then the parameter $\lambda=Ef(X)$ can be estimated using either IID Monte Carlo (IIDMC) or Markov chain Monte Carlo (MCMC) methods. In this talk we describe a new monte carlo procedure called RSMC (Regenerative sequence monte carlo) for this problem and show it works for the case whether $\pi$ is proper, i.e, $\pi(S) <\infty$ or not proper, i.e, $\pi(S)=\infty$. We illustrate this with estimating a convergent infinite series and an integral over $R$ with respect to Lebesgue measure. We also provide a confidence interval for $\lambda$.