THIS REPORT HAS BEEN WITHDRAWN

Xiaofei Zhou

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2015-228
December 4, 2015

We apply coarse-to-fine MCMC to perform Bayesian inference for a seismic monitoring system. While traditional MCMC has difficulty moving between local optima, by applying coarse-to-fine MCMC, we can adjust the resolution of the model and this allows the state to jump between different optima more easily. It is quite similar to simulated annealing. We will use a 1D model as an example, and then compare traditional MCMC with coarse-to-fine MCMC and discuss the scaling behavior.

Advisor: Stuart J. Russell

Author Comments: see http://www2.eecs.berkeley.edu/Pubs/TechRpts/2015/EECS-2015-252.html