Abstract
We present a novel representation learning method for downstream tasks likeanomaly detection, unsupervised classification, and similarity searches inhigh-energy data sets. This enabled the discovery of a new extragalactic fastX-ray transient (FXT) in Chandra archival data, XRT 200515, aneedle-in-the-haystack event and the first Chandra FXT of its kind. Recentserendipitous discoveries in X-ray astronomy, including FXTs from binaryneutron star mergers and an extragalactic planetary transit candidate,highlight the need for systematic transient searches in X-ray archives. Weintroduce new event file representations, E-t maps and E-t-dt cubes, thateffectively encode both temporal and spectral information, enabling theseamless application of machine learning to variable-length event file timeseries. Our unsupervised learning approach employs PCA or sparse autoencodersto extract low-dimensional, informative features from these datarepresentations, followed by clustering in the embedding space with DBSCAN. Newtransients are identified within transient-dominant clusters or throughnearest-neighbour searches around known transients, producing a catalogue of3559 candidates (3447 flares and 112 dips). XRT 200515 exhibits unique temporaland spectral variability, including an intense, hard <10s initial burst,followed by spectral softening in an ~800s oscillating tail. We interpret XRT200515 as either the first giant magnetar flare observed at low X-ray energiesor the first extragalactic Type I X-ray burst from a faint, previously unknownlow-mass X-ray binary in the LMC. Our method extends to data sets from otherobservatories such as XMM-Newton, Swift-XRT, eROSITA, Einstein Probe, andupcoming missions like AXIS.