Artificial Neural Networks as a Tool to Classify the 2FGL Unassociated Sources

The Fermi Large Area Telescope Second Source Catalog (2FGL) lists positional, spectral, and temporal properties for 1873 gamma-ray sources detected during the first 24 months of operation in the 100 MeV to 300 GeV energy band. Approximately 30% of these sources remain "unassociated", i.e. do not have any plausible known counterpart. The improved statistics from the LAT have enabled us to characterize each source with remarkable detail. We report on the use of Artificial Neural Networks (ANN) as a very promising method for understanding the nature of Fermi-LAT unassociated sources. This technique uses identified objects as a training sample, learning to distinguish each source class on the basis of parameters that describe its gamma-ray properties. By applying the algorithm to unknown objects, such as the unassociated sources, it is possible to quantify their probability of belonging to a specific class. We will present the ANN algorithm and discuss its application for classifying the 2FGL unassociated sources, its performance, and the advantages and disadvantages as compared with other classification schemes.