Speaker
Description
The Bayesian data analysis approach combines prior knowledge and observed data to derive information about the parameters of a model. Typically, numerical sampling methods are required for performing Bayesian inference due to the complexity of the models and the high-dimensional parameter spaces involved, as is usually also the case in particle physics applications. Markov Chain Monte Carlo (MCMC) methods are commonly used to generate samples that approximate the posterior distribution. Machine learning techniques have the potential to enhance MCMC methods by improving the exploration of complex parameter spaces, leading to more accurate results. This talk focuses on normalizing flow models, which allow to transform a complex distribution into a simpler one, thereby improving the sampling efficiency of MCMC algorithms. This presentation introduces an implementation of a normalizing flow enhanced MCMC ensemble algorithm currently being integrated into the Bayesian Analysis Toolkit (BAT.jl). Initial studies on the performance of this new algorithm are presented.
