Fast Hybrid Tempered Ensemble Transform Filter for Bayesian Elliptical Problems

Learn about advanced Bayesian inverse problem solving techniques in this one-hour seminar presentation from Dr. Jana de Wiljes of Universität Potsdam. Explore the challenges of working with partial and noisy data in high-dimensional and nonlinear settings, and discover how ensemble Kalman filtering provides robust estimations through Gaussian approximations. Examine the limitations of traditional methods when dealing with non-Gaussian posterior distributions and understand the development of the tempered ensemble transform particle filter as an adaptive sequential Monte Carlo method. Delve into the introduction of an entropy-inspired regularization factor that reduces computational costs through Sinkhorn iterations, and learn how incorporating an ensemble Kalman filtering proposal step creates a more robust hybrid approach for solving complex Bayesian elliptical problems.