Abstract: We present a search for dark matter (DM) annihilating subhalos of the Milky Way halo among the Fermi Large Area Telescope (LAT) unassociated sources. For this purpose, we construct the first statistical model of the unassociated sources at latitudes above 10 degrees, combining potential DM subhalos with Galactic and extragalactic astrophysical components. The distributions of astrophysical sources are constructed based on associated sources, while the DM subhalo distribution is derived from Monte Carlo simulations. We account for differences between associated and unassociated source distributions using a model with covariate and prior probability shifts, which are particular cases of more general dataset shifts. This approach is based on quantification learning, advancing beyond previous classify-and-count strategies by providing a well-defined statistical interpretation of the potential contribution from a DM subhalo population. For the bb & strns; annihilation channel and DM masses from 10 GeV to 1 TeV, we find no significant contribution from DM subhalos, and therefore derive 95% confidence upper limits on the annihilation cross section. Our analysis yields limits consistent with previous classify-and-count approaches, while the underlying generative model provides a more robust statistical framework, opening new avenues for population studies of Fermi-LAT sources and, more generally, for searches of anomalies, such as a new class of sources in addition to the known classes of sources, in presence of statistical and systematic uncertainties.

Abstract: Neural simulation-based inference (NSBI) is a powerful class of machine-learning-based methods for statistical inference that naturally handles high-dimensional parameter estimation without the need to bin data into low-dimensional summary histograms. Such methods are promising for a range of measurements, including at the Large Hadron Collider, where no single observable may be optimal to scan over the entire theoretical phase space under consideration, or where binning data into histograms could result in a loss of sensitivity. This work develops a NSBI framework for statistical inference, using neural networks to estimate probability density ratios, which enables the application to a full-scale analysis. It incorporates a large number of systematic uncertainties, quantifies the uncertainty due to the finite number of events in training samples, develops a method to construct confidence intervals, and demonstrates a series of intermediate diagnostic checks that can be performed to validate the robustness of the method. As an example, the power and feasibility of the method are assessed on simulated data for a simplified version of an off-shell Higgs boson couplings measurement in the four-lepton final states. This approach represents an extension to the standard statistical methodology used by the experiments at the Large Hadron Collider, and can benefit many physics analyses.