We will present what we call the Markov Field Monte Carlo, a method for sampling Markov Fields, known most widely as graphical models: creatures ubiquitous in many branches of science.
In the talk we will provide a brief introduction to graphical models and a process known as renormalization. We use a method called Fast Marginalization to efficiently simplify a given graphical model and use resulting structure to sample the original model. Application of this method to a toy problem, the Ising model in one and two dimensions, will be shown in detail. Time permitting, we will discuss generalizations of this approach that give a more complete theoretical picture of our work.
The talk will be accessible to anyone with rudimentary understanding of Markov Chain Monte Carlo; familiarity with the Ising model and graphs is helpful.