Pedro Alexandre C. Cunha
IA / FCUP

Abstract
Recent applications of machine learning techniques in Astronomy have been data-driven with less focus on smaller data sets with high scientific potential. Here, we explore the application of the few-shot learning approach to identifying type II quasars (QSO2) using photometric data. The training data comprises QSO2 from the SDSS BOSS and galaxies from the SDSS at 1 ≤ z ≤ 4. We present the A.M.E.L.I.A. pipeline which uses machine-transfer-learning based approach for the selection of QSO2 candidates using the SDSS optical broadband magnitudes and WISE photometry and takes advantage of decision-trees, distance-based and deep learning methods to build a stronger single classifier using the generalised stacking technique. We tested the performance of A.M.E.L.I.A. in a supervised mode and obtained high precision and recall with an F1-score ≥ 0.9, for a binary classification setup. The pipeline was used in semi-supervised mode to select QSO2 candidates from a sample of SDSS spectroscopic classified galaxies. We discuss the nature of our QSO2 candidates, including a sub-population of [NeV]λ3426 emitters at z ∼ 1.1, which are highly likely to contain obscured AGNs. Our results suggest that the application of few-shot learning in Astronomy can help increase the number of sources with low statistics and that the correlations found by the machine learning models are robust to transfer-learning.

2023 March 30, 13:30

IA/U.Porto
Centro de Astrofísica da Universidade do Porto (Auditorium)
Rua das Estrelas, 4150-762 Porto