Monte Carlo sampling techniques are often either inefficient or practically impossible when sampling from high-dimensional and complicated distributions. In such situations, Markov chain Monte Carlo techniques are popularly employed. However, owing to the introduction of correlated sampling, special care must be taken to ensure ergodicity of the underlying process. Dr Dootika will motivate the need for Markov chain Monte Carlo (MCMC) and explain popular algorithms that yield theoretically well-behaved Markov chains. The popular Metropolis-Hastings algorithm will be given special attention and some variants of the algorithm will be discussed. Further, Dr Dootika will touch upon the practical difficulties of implementing MCMC algorithms in the modern world of big data.
ABOUT THE SPEAKER
Dr Dootika Vats is an Associate Professor in the Department of Mathematics and Statistics at the Indian Institute of Technology, Kanpur. Her research interests are Markov chain Monte Carlo, output analysis for stochastic simulation. Recently, she has been interested in stochastic optimization algorithms. She is the recipient of ANRF MATRICS, “Proximal Markov Chain Monte Carlo for Constrained and Nonsmooth Targets."
