Superluminous supernova search with PineForest

TL;DR

This paper demonstrates how active learning with the PineForest algorithm can efficiently identify rare superluminous supernova candidates from large astronomical data sets, leveraging prior confirmed sources to improve discovery rates.

Contribution

It introduces a novel application of active learning with PineForest for supernova searches, showing improved efficiency using prior confirmed sources in large data sets.

Findings

Identified 8 SLSN candidates from 120 scrutinized objects.

Discovered 2 new SLSN candidates not previously reported.

Showed the effectiveness of prior confirmed sources in active learning.

Abstract

The advent of large astronomical surveys has made available large and complex data sets. However, the process of discovery and interpretation of each potentially new astronomical source is, many times, still handcrafted. In this context, machine learning algorithms have emerged as a powerful tool to mine large data sets and lower the burden on the domain expert. Active learning strategies are specially good in this task. In this report, we used the PineForest algorithm to search for superluminous supernova (SLSN) candidates in the Zwicky Transient Facility. We showcase how the use of previously confirmed sources can provide important information to boost the convergence of the active learning algorithm. Starting from a data set of $\sim$14 million objects, and using 8 previously confirmed SLSN light curves as priors, we scrutinized 120 candidates and found 8 SLSN candidates, 2 of which have not been reported before (AT 2018moa and AT 2018mob). These results demonstrate how existing spectroscopic samples can be used to improve the efficiency of active learning strategies in searching for rare astronomical sources.