In this work, Machine Learning (ML) methods are used to efficiently identify the unassociated sources and the Blazar Candidate of Uncertain types (BCUs) in the Fermi-LAT Third Source Catalog (3FGL). The aims are twofold: (1) to distinguish the Active Galactic Nuclei (AGNs) from others (non-AGNs) in the unassociated sources; (2) to identify BCUs into BL Lacertae objects (BL Lacs) or Flat Spectrum Radio Quasars (FSRQs). Two dimensional reduction methods are presented to decrease computational complexity, where Random Forest (RF), Multilayer Perceptron (MLP) and Generative Adversarial Nets (GAN) are trained as individual models. In order to achieve better performance, the ensemble technique is further explored. It is also demonstrated that grid search method is of help to choose the hyper-parameters of models and decide the final predictor, by which we have identified 748 AGNs out of 1010 unassociated sources, with an accuracy of 97.04%. Within the 573 BCUs, 326 have been identified as BL Lacs and 247 as FSRQs, with an accuracy of 92.13%.

在这项工作中，机器学习（ML）方法用于有效地识别费米中未关联的来源和不确定类型的Blazar候选对象（BCU）-LAT第三来源目录（3FGL）。目的是双重的：（1）区分活动银河核（AGN）与未关联源中的其他（非AGN）；（2）将BCU识别为BL Lacertae对象（BL Lacs）或Flat Spectrum Radio Quasars（FSRQ）。提出了二维减少方法以减少计算复杂性，其中随机森林（RF），多层感知器（MLP）和生成对抗网络（GAN）被作为单个模型进行训练。为了获得更好的性能，进一步探索了集成技术。还证明了网格搜索方法有助于选择模型的超参数并确定最终的预测变量，据此我们从1010个未关联的源中识别出748个AGN，准确度为97.04％。在573个BCU中，有326个被标识为BL Lacs，有247个被标识为FSRQ，