Sampling through Exploration Exploitation in Markov Chain Monte Carlo
Offered By: Institut des Hautes Etudes Scientifiques (IHES) via YouTube
Course Description
Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore an innovative Markov chain Monte Carlo algorithm called Ex2MCMC in this 35-minute conference talk by Eric Moulines from Ecole Polytechnique. Delve into the development of this massively parallelizable and computationally efficient method that combines multiple global proposals with mobile moves. Learn about the algorithm's V-uniform geometric ergodicity under realistic conditions and understand the explicit bounds on mixing rates that demonstrate improvements due to multiple global moves. Discover how Ex2MCMC allows for fine-tuning of exploitation (local moves) and exploration (global moves) through a novel approach to proposing dependent global moves. Examine the adaptive scheme, FlEx2MCMC, which learns the distribution of global trains using normalizing flows. Gain insights into the efficiency of Ex2MCMC and its adaptive versions through various classical sampling benchmarks, and see how these algorithms enhance the quality of sampling GANs as energy-based models.
Syllabus
Eric Moulines - Sampling through Exploration Exploitation
Taught by
Institut des Hautes Etudes Scientifiques (IHES)
Related Courses
Bayesian Computational StatisticsIllinois Institute of Technology via Coursera Mathematical understanding of uncertainty
Seoul National University via edX Probabilistic Graphical Models 2: Inference
Stanford University via Coursera Action Matching: Learning Stochastic Dynamics from Samples
Generative Memory Lab via YouTube Mathematical Foundation of Various MCMC Methods
Institut des Hautes Etudes Scientifiques (IHES) via YouTube