The unicic module implements the Feldman-Cousins “unified” confidence interval construction. Such calculations can be lengthy, particularly when the confidence interval is a function of more than one parameter. Past work attacks this computing challenge by examining how to scale the grid search to HPC or how to efficiently span the parameter space with Gaussian processes. This work attempts a third orthogonal approach of accelerating with GPU. In principle any two or all three may be combined.
The calculation is summarized here. See the original F-C paper for motivations and details.