Josh G. Weston

We present a convolutional neural network (CNN) for use in the real–bogus classification of transient detections made by the Asteroid Terrestrial-impact Last Alert System (ATLAS) and subsequent efforts to improve performance since initial development. In transient detection surveys, the number of alerts made outstrips the capacity for human scanning, necessitating the use of machine learning aids to reduce the number of false positives presented to annotators. We take a sample of recently annotated data from each of the three operating ATLAS telescope with \\∼\340 000 real (known transients) and \\∼\1030 000 bogus detections per model. We retrained the CNN architecture with these data specific to each ATLAS unit, achieving a median false positive rate (FPR) of 0.72 per cent for a 1.00 per cent missed detection rate. Further investigations indicate that if we reduce the input image size it results in increased FPR. Finally architecture adjustments and comparisons to contemporary CNNs indicate that our retrained classifier is providing an optimal FPR. We conclude that the periodic retraining and readjustment of classification models on survey data can yield significant improvements as data drift arising from changes in the optical and detector performance can lead to new features in the model and subsequent deteriorations in performance.

The upcoming Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) will enable astronomers to discover rare and distant astrophysical transients. Host-galaxy association is crucial for selecting the most scientifically interesting transients for follow-up. LSST deep drilling field (DDF) observations will detect distant transients occurring in galaxies below the detection limits of most all-sky catalogs. Here, we investigate the use of preexisting, field-specific catalogs for host identification in the DDFs and a ranking of their usefulness. We have compiled a database of 70 deep catalogs that overlap with the Rubin DDFs and constructed thin catalogs to be homogenized and combined for transient-host matching. A systematic ranking of their utility is discussed and applied based on the inclusion of information such as spectroscopic redshifts and morphological information. Utilizing this data against a Dark Energy Survey sample of supernovae with pre-identified hosts in the XMM-Large Scale Structure and the Extended Chandra Deep Field-South fields, we evaluate different methods for transient-host association in terms of both accuracy and processing speed. We also apply light data-cleaning techniques to identify and remove contaminants within our associations, such as diffraction spikes and blended galaxies where the correct host cannot be determined with confidence. We use a lightweight machine learning approach in the form of extreme gradient boosting to generate confidence scores in our contaminant selections and associated metrics. Finally, we discuss the computational expense of implementation within the LSST transient alert brokers, which will require efficient, fast-paced processing to handle the large stream of survey data.