Indomitable segmentation and pooling: A new perspective on clairvoyant fusion
Wednesday, October 24, 2018 - 9:00am - 9:50am
The universe of solutions to composite hypothesis testing problems was greatly inflated with the development of clairvoyant fusion (CF) methods. This formalism encompasses several older methods and allows closed form solutions to problems not amenable to the classical approaches. However, the CF choices are so expansive that often guidelines seem necessary. For example, when unknown values of model parameters are associated with both null and signal hypotheses, a preferred order for dual fusing has never been established. The process is not generally commutative. Here it is shown that indomitable (admissible) segmentation followed by pooling is equivalent to a constrained fusion method and provides a natural guideline. The constraints are automatically satisfied in unitary fusion, thus justifying most prior work. Furthermore, they restrict dual fusion to a manageable set of options, and make it commutative to boot.