|
Abstract:
|
The detection of bumps in energy spectra, the identification of unexpected features in astronomical images, and the ability to distinguish between a new emission and the signal of known sources mimicking it, are problems of crucial importance in both physics and astronomy. In this work, we show that all these challenges can be formulated in statistical terms as a test of hypothesis where a nuisance parameter is present only under the alternative, and we discuss a computational solution to perform inference in this setting. The methodology proposed is highly generalizable and combines elements of extreme value theory, graph theory and simulations methods to achieve ease of implementation and computational efficiency under stringent significance requirements.
|
