Title: Adaptive Schemes for Sampling in Infinite Dimensions
Speaker: Sreekar Vadlamani, TIFR-CAM, Bengaluru
Time and Date: 10 AM to 11 AM on 25th November, 2019 (Monday)
Venue: Seminar Room (2nd Floor, IEOR Building)
Abstract:
Latent Gaussian processes are widely applied in many fields like statistics, inverse problems, and machine learning. A popular method for inference is through the posterior distribution, which is typically carried out by Markov Chain Monte Carlo (MCMC) algorithms. However, the infinite dimensional framework creates certain technical hurdles which need to be addressed with care. Taking cue from recent developments in adaptive Metropolis adjusted MCMC algorithms, we propose a family of proposals to sample from "good" infinite dimensional measures. We discuss the relevant issues concerned with convergence of our proposed schemes, and also demonstrate their efficiency via standard computational examples.
Speaker Bio:
Sreekar Vadlamani is from TIFR-Centre for Applicable Mathematics. He obtained his Ph.D from Technion. He has also spent sometime at Stanford and at Technion as postdoc prior to joining TIFR-CAM. His research interests are in Geometric aspects of random fields, Sampling in infinite dimensions and Random graphs.