Speaker:Dr. Christian Shelton

Multi-Fidelity Concentric MCMC

Markov chain Monte Carlo (MCMC) is a workhorse of scientific and engineering computation because it requires only the ability to evaluate the model's likelihood. However, this is also its weakness because it has no additional knowledge to guide the sampling, thus often making the speed at which it generates effectively new samples slow. Multi-Fidelity Concentric MCMC (MC-cubed) speeds up MCMC on scientific and engineering inference problems where the underlying system is a simulation with tunable fidelity. It employs recursive Markov chains at coarser fidelities to guide the proposal distribution at higher fidelities. The result is an MCMC method for the highest (true) fidelity generates more effective samples per total computation time, even considering the extra cost of the coarser chains. The computational savings are demonstrated on hydrology and cosmology parameter estimation problems.

Biography

Dr. Shelton is a Professor of Computer Science at the University of California at Riverside and a member of UCR's Data Science Center. He has been on the faculty since 2003. His research interests are in statistical approaches in artificial intelligence, with a focus on machine learning and dynamic systems. He has applied his work to areas ranging from astronomy to sociology to medical informatics.
Dr. Shelton received his B.S. in Computer Science from Stanford University in 1996. He then obtained his Ph.D. from MIT in 2001 and returned to Stanford from 2001 to 2003 as a post-doctoral scholar. He spent six months in 2003 and 2004 as a visiting faculty member at Intel Research and the 2012-13 academic year as a visiting researcher at Children's Hospital Los Angeles. He has been the Managing Editor of the Journal of Machine Learning Research and on the Editorial Board of the Journal of Artificial Intelligence Research; he currently serves on the Editorial Board for the Journal of Machine Learning Research.