Friday, January 22, 2:50 - 4:05 PM, Asif Jamil Chowdhury, University of South Carolina Abstract: Markov Chain Monte Carlo (MCMC) has become the main computational workhorse in scientific computing for solving statistical inverse problems. It is difficult however to use MCMC algorithms when the likelihood function is computational expensive to evaluate.Here, a novel Metropolis-Hastings algorithm is proposed to sample from posterior distributions corresponding to computationally expensive simulations. The main innovation is emulating the likelihood function using Gaussian processes. The proposed emulator is constructed on the fly as the MCMC simulation evolves and adapted based on the uncertainty in the acceptance rate. The algorithm is tested on a number of benchmark problems where it is shown that it significantly reduces the number of forward simulations. Bio: Asif Jamil Chowdhury is a graduate student in the department of Computer Science and Engineering at University of South Carolina. His supervisor is Dr. Gabriel Terejanu. His primary research interests lie in the field of uncertainty quantification and model validation. At present he is working on the use of Gaussian Processes in Bayesian optimization and Markov Chain Monte Carlo methods. Before starting his graduate studies he worked as software developer for seven years. This seminar is open to anyone who is interested, not just students enrolled in the CSCE 791 class. Please consider attending.