The traditional method of Gravitational Wave (GW) detection is Matched Filtering that was used for the first GW detection by aLIGO in 2015. The method works by matching the observation data sample with a set of templates of known GW waveforms.
Iterating through all the templates for relatively complex GW signals, for instance those from eccentric sources, increases the overall computational cost and time complexity.
In recent years, Machine Learning techniques have been probed as a solution to this problem. In this short paper, we present a new Convolutional Neural Network model for detection of GW signals from Neutron Star$-$Black Hole (NSBH) binaries in Gaussian random noise. We use NSBH signals simulated using IMRPhenomNSBH LALsuite waveform approximant for training the model. We then compare the model detection sensitivities for three different training strategies obtained by combining Uniform and Non-uniform Signal-to-Noise distribution in the training dataset with the Curriculum Learning training methodology. We find that two of the training strategies perform considerably better than the other one for all test datasets considered.