Using machine learning and the Hough transform to search for gravitational waves due to r-mode emission by isolated neutron stars

We present two new methods to search for gravitational waves (GWs) from isolated neutron stars lasting O(hours-days), and the results of three searches of real LIGO/Virgo data. One method is a generalization of a method to detect continuous waves (CWs) that looks for signals that follow a specific model; the other uses machine learning to detect signals that deviate from this model. In two of our searches, we tried to find a remnant of the first-ever binary neutron star (BNS) merger, GW170817; in the third search, we looked everywhere in the sky for any kind of long duration transient signal in LIGO's third observing run. For the first method developed, the Generalized Frequency Hough, we present empirical and theoretical estimates of its sensitivity, an extensive follow-up procedure of GW candidates, the parameter space to which it is sensitive, and computational constraints when using it. For the second method, an application of convolutional neural networks, we demonstrate that its sensitivity is comparable to that of first method but is much faster to apply and can find signals that the first method cannot. Estimates of rates and sensitivities in future detectors, as well as implications of a GW detection from a neutron star, are also discussed. Additionally we explore the prospects of using a two-dimensional Fast Fourier Transform to improve the quality of time/frequency images to be used by either method.