Ponente
Descripción
About half of the Fermi-LAT gamma-ray sources within plus or minus 10 degrees in latitude from the Galactic plane are unassociated. Interestingly, the distribution of the Galactic unassociated sources as a function of spectral parameters is different from the distributions of known classes of Galactic sources, such as pulsars, supernova remnants, and pulsar wind nebulae. This difference in distributions is not only puzzling from the point of view of known classes of gamma-ray emitters, it is also a challenge for standard machine learning classification, where it is assumed that the distributions of training and target datasets are the same. In this talk I will discuss how machine learning can be used in modeling Galactic unassociated sources taking into account dataset shifts between training and target distributions and what such models can tell us about the physical nature of low-latitude unassociated sources.
