Acceleration of Bayesian model based data analysis

A general solution for dealing with parameter estimation in a rigorous way is Bayesian data analysis. This analysis allows estimation of specific parameters and their uncertainties for non-linear inverse problems in a strictly mathematical way. Though advantageous, it is computationally intensive with long processing times and therefore, with exceptions such as the Kalman filter for linear systems with Gaussian noise, its full analysis is not commonly used for real-time applications. For non-linear problems, there are many suboptimal Bayesian online algorithms using different approaches which all introduce some kind of approximation. Nevertheless, an approach avoiding approximations to improve inference is desired. An acceleration of Bayesian analysis for inverse problems using reconfigurable hardware or multi-core processing is necessary and could prove valuable for scientific inference. This paper describes the acceleration of the Dispersion Interferometer (DI) for the Wendelstein 7-X (W7-X) magnetic confinement device as a proof of principle of the Bayesian model based data analysis.