Sparse grid nested sampling for model selection in eddy-current testing

Model selection is a common problem that one can run into in non-destructive evaluations. Without any further information, uniform distributions can be used as the prior models. Among the methods dedicated to model evidence estimation, Nested Sampling (NS) is one of the most efficient one. Compared to traditional Monte-Carlo methods, it offers a good compromise between the computational cost and the ability to manage complicated objective functions. In the present work, we use an accelerated NS method. The acceleration benefits from the existing points in the database and narrows down the parameter search space at the initialisation.