ulisse: A tool for one-shot sky exploration and its application for detection of active galactic nuclei

Abstract

Context. Modern sky surveys are producing ever larger amounts of observational data, which makes the application of classical approaches for the classification and analysis of objects challenging and time-consuming. However, this issue may be significantly mitigated by the application of automatic machine and deep learning methods. Aims. We propose ULISSE, a new deep learning tool that, starting from a single prototype object, is capable of identifying objects sharing the same morphological and photometric properties, and hence of creating a list of candidate sosia. In this work, we focus on applying our method to the detection of AGN candidates in a Sloan Digital Sky Survey galaxy sample, since the identification and classification of Active Galactic Nuclei (AGN) in the optical band still remains a challenging task in extragalactic astronomy. Methods. Intended for the initial exploration of large sky surveys, ULISSE directly uses features extracted from the ImageNet dataset to perform a similarity search. The method is capable of rapidly identifying a list of candidates, starting from only a single image of a given prototype, without the need for any time-consuming neural network training. Results. Our experiments show ULISSE is able to identify AGN candidates based on a combination of host galaxy morphology, color and the presence of a central nuclear source, with a retrieval efficiency ranging from 21 % to 65 % (including composite sources) depending on the prototype, where the random guess baseline is 12 %. We find ULISSE to be most effective in retrieving AGN in early-type host galaxies, as opposed to prototypes with spiral-or late-type properties. Conclusions. Based on the results described in this work, ULISSE can be a promising tool for selecting different types of astrophysical objects in current and future wide-field surveys (e.g. Euclid, LSST etc.) that target millions of sources every single night.

The LSST AGN Data Challenge: Selection Methods

“…In this section, we summarize four out of the five "solutions" submitted to the DC (since one using CNNs and transfer learning has been already been presented by Doorenbos et al 2022). One submission used a non-ML method that extended the traditional approach based on color-color diagrams by adding magnitude standard deviation and the coefficients of correlation between light curves in two wave bands.…”

“…In addition, we find that 64 × 64 pixel image cutouts are not sufficient for extracting pixel-level information. However, in the parallel effort done on low-redshift/low-luminosity AGNs (Doorenbos et al 2022), it was found that images of resolution 224 × 224 pixels can be used to directly extract meaningful information that can help in disentangling AGNs from non-AGN hosts. Moreover, the quality of images in all six bands that will be provided by LSST will allow us to develop DL models for separating AGNs from galaxies in a highly efficient manner at the cost of additional computing power.…”

“…One advantage of DL is that it allows for simultaneous training of models with multiple inputs (e.g., catalog features + images) and for multiple outputs (e.g., regression and classification, Chollet et al 2015). An important property of DL is the application of transfer learning (Tan et al 2018), i.e., a model trained for one task is repurposed on a second, related task (Doorenbos et al 2022). For the application of DL in astronomy, we refer to a review by Smith & Geach (2022).…”

“…The implementation of difference image analysis (DIA) in the LSST reduction pipelines will completely revolutionize the detection of AGNs through variability by removing the contribution from the host galaxy (Zebrun et al 2001;Bramich 2008;Kozłowski et al 2010aKozłowski et al , 2016. Finally, the addition of high-resolution images is expected to considerably increase the performance of the selection methods (Doorenbos et al 2022).…”

“…Five proposed solutions were submitted to the DC: one using a classical approach, and four applying ML-based AGN selection. We will present the solution using the classical approach and three ML-based solutions, while the one remaining ML-based solution is addressed by Doorenbos et al (2022).…”