For more information about this meeting, contact Kris Jenssen, Yuxi Zheng.
|Title:||Using Markov Chains to Calibrate High Dimensional Models|
|Seminar:||CCMA Luncheon Seminar|
|Speaker:||John C. Liechty, Smeal College of Business, Penn State University|
|The ability to construct a stationary or target distribution, that is consistent with a particular statistical model (predictive model + error) of interest and allows for the possibility of calibrating this model by creating a discrete-time, general state-space Markov chain over the space of possible parameter values. The Hastings-Metropolis algorithm, which originally came from Statistical Physics, offers a way of constructing a Markov Chain that has the desired 'target' distribution as its unique stationary distribution - a calibration effort which is known as Markov Chain Monte Carlo. I will motivate the theory by reviewing some relevant discrete-time Markov Chain theory and then give a few illustrative examples from the Bayesian Statistical literature.|
Room Reservation Information
|Date:||02 / 26 / 2010|
|Time:||12:00pm - 01:30pm|