Quasi-Anomalous Gravitational-Wave Detection with Recurrent Autoencoders

Detection of gravitational wave (GW) signals in laser interferometers relies on having well modeled templates of the GW emission. We present a method of anomaly detection techniques based on deep recurrent autoencoders to the enhance the search region to potential, unmodelled transients. We use a semi-supervised strategy dubbed Quasi Anomalous Knowledge (QUAK) which provides a weak distinction between classes at training time. While the semi-supervised nature of the problem comes with a cost in terms of accuracy as compared to supervised techniques, there is a qualitative advantage in generalizing experimental sensitivity beyond pre-computed signal templates. We construct a low-dimensional embedded space using the QUAK method which captures the physical signatures of distinct signals on each axis of the space. By introducing alternative signal priors that capture some of the salient features of gravitational-wave signatures, we allow for the recovery of sensitivity even when an unmodelled anomaly is encountered. We show that regions of the QUAK space can identify binaries, detector glitches and also search a variety of hypothesized astrophysical sources that may emit GWs in the LIGO frequency band, including core-collapse supernovae and other stochastic sources.