Hybrid deep learning for blazar classification and correlation search with neutrinos

Since the successful discovery of neutrinos, their origin is still a mystery until the association with TXS 0506+056. More messengers are achievable to study the intrinsic mechanism of extreme phenomena. Recently, Fermi-LAT Collaboration published the fourth catalogue of active galactic nuclei (4LAC), together with Data Release 2 later, including totally 3148 sources. The majority of these objects are blazars, which are classified into flat-spectrum radio quasars (FSRQs), BL Lac-type objects (BLLs), and blazars of uncertain type (BCUs) according to their optical observation feature. The BCUs take up to 38.2 per cent of total, whose classification is quite challenging and manpower consuming. However with the remarkable advances of technology, deep learning has been widely applied in astronomy. In this work, we take the advantage of 11 machine learning algorithms plus the convolutional neural network (CNN)-based deep learning algorithm to classify BCUs based on 10 parameters and the broad-band spectral energy distribution of each object obtained with vou-blazars. On average, this method has impressive performance, reaching above 95 per cent of balanced accuracy for the training sample, best among the studies so far. We correlate the IceCube neutrinos and blazars in 4LAC, found a few possible associations. With the machine learning prediction, we later elaborate the association of these BCUs with neutrinos samples and find out most of the BCUs associated with neutrinos are with low synchrotron peak frequency, which may be due to the energy distribution of high-energy particles. We propose neutrinos might be another feature for objects classification in the future.