Transient-optimized real-bogus classification with Bayesian convolutional neural networks – sifting the GOTO candidate stream

Abstract

Large-scale sky surveys have played a transformative role in our understanding of astrophysical transients, only made possible by increasingly powerful machine learning-based filtering to accurately sift through the vast quantities of incoming data generated. In this paper, we present a new real-bogus classifier based on a Bayesian convolutional neural network that provides nuanced, uncertainty-aware classification of transient candidates in difference imaging, and demonstrate its application to the datastream from the GOTO wide-field optical survey. Not only are candidates assigned a well-calibrated probability of being real, but also an associated confidence that can be used to prioritise human vetting efforts and inform future model optimisation via active learning. To fully realise the potential of this architecture, we present a fully-automated training set generation method which requires no human labelling, incorporating a novel data-driven augmentation method to significantly improve the recovery of faint and nuclear transient sources. We achieve competitive classification accuracy (FPR and FNR both below 1%) compared against classifiers trained with fully human-labelled datasets, whilst being significantly quicker and less labour-intensive to build. This data-driven approach is uniquely scalable to the upcoming challenges and data needs of next-generation transient surveys. We make our data generation and model training codes available to the community.

The Gravitational-wave Optical Transient Observer (GOTO): prototype performance and prospects for transient science

“…For a fixed false positive rate of 1 per cent, the newlyimplemented classifier achieved a 1.5 per cent false negative rate on a held-out test set, and reached a ∼97 per cent recovery rate when evaluated on a benchmark dataset of real observations of confirmed transients. The CNN model and the automated data-generation techniques are described fully in Killestein et al (2021).…”

The Gravitational-wave Optical Transient Observer (GOTO)

“…Candidates are classified using a convolutional neural network, which assigns each source a score between "real" (1) and "bogus" (0). 16,17 High-scoring sources are presented for human vetting through a web interface called the GOTO Marshall (see Figure 4). Collaboration members can check each candidate and flag them either as potential astrophysical sources or junk detections, based on historic light curves, image stamps and cross-matching with astrophysical catalogues.…”

Light curve classification with recurrent neural networks for GOTO: dealing with imbalanced data

“…as a result of poor subtraction from difference imaging) or a cosmic ray. The real-bogus classification problem has been well studied and applied to surveys, and is becoming a standard part of automated discovery pipelines (Killestein et al 2021;Mong et al 2020;Duev et al 2019;Lin et al 2018;Gieseke et al 2017;Wright et al 2015;Brink et al 2013).…”

“…While not in gravitational wave follow-up mode, GOTO conducts an all-sky survey to search for transient and variable sources. A realbogus classifier (Killestein et al 2021) first identifies astrophysically real detections and 'bogus' subtraction artefacts from difference images. Detections that are identified as real by the real-bogus classifier are then passed on to the GOTO Marshall (Lyman et al in prep) , a web-based interface for GOTO observers to vet, search, and trigger follow-up observations of new discoveries.…”