Deep learning for the selection of Young Stellar Object candidates from IR surveys

The robust identification of YSOs is an important step for characterizing star-forming regions. Such classification is commonly performed with infrared suveys using straight cuts in CMD diagrams. However, Machine Learning algorithms may outperform these methods with adaptive and non-linear separations in any number of dimensions. In this paper we present our methodology to implement a supervised deep neural network for YSO classification with various datasets built from well-known regions. We detail the tuning of the network parameters, taking into account the specificities of this classification. Then we focus on the reliability of the classification and address difficulties due to the strong dilution of YSOs against contaminants.