Large Scale Bayesian Data Analysis for Nuclear Fusion Experiments

There is currently a paradigm shift taking place in the field of scientific methodology. Methods for the extraction of underlying physics from observations and the falsification/confirmation of scientific hypothesis are undergoing a significant change through the use of a generic approach to inference from observations: so called 'Bayesian' Probability Theory. The first part of this paper will outline and exemplify how this method is changing data analysis in nuclear fusion: How all uncertainties (systematic, statistical and model uncertainties) can be treated in a unified way, and how data analysis methods can be understood and unified through probability theory. The practical advantage here for nuclear fusion experiments is the possibility to utilise this method for a more comprehensive understanding of the internal state of fusion plasmas as inferred from measurements from multiple heterogeneous diagnostics. The second part of the paper will discuss architectural issues relating to the very complex analysis systems that might emerge from a systematic application of this method in large scientific experiments.