Speaker
Description
Gravitational-wave (GW) astronomy has progressed steadily since the first detection of a signal from a binary black-hole coalescence in 2015, and in parallel, machine-learning techniques have become increasingly valuable allies in this field, enabling more effective searches for different signal types and improving characterization of detector noise. Although all confirmed detections to date originate from Compact Binary Coalescences (CBC), Core-Collapse Supernovae (CCSNe) remain a promising source class to be detected. They are expected to produce stochastic burst-like signals for which matched-filtering approaches are inefficient. The detection challenge is further aggravated when one accounts for the duty cycle of the detector network, since there are periods in which only a single interferometer is operating; a galactic CCSN occurring during such a period would be recorded by a lone instrument, and multi-detector correlation would be impossible. Motivated by these limitations, the present work aims to develop a neural network for binary classification capable of deciding whether an event is signal or noise, using event lists extracted from the coherent WaveBurst (cWB) pipeline. To produce the training dataset, simulated CCSN waveforms were injected into real LIGO background data, generating a catalog that mixes simulated signals and typical glitches; this catalog is then used for supervised learning and model evaluation. In light of this, here we present the current status of the project and some preliminary results, in which the classifier shows promise in classifying the search background and increasing contrast on genuine CCSN candidates, which could aid follow-up prioritization and improve the effective reliability of single-interferometer searches. Overall, this effort seeks to contribute to expanding the detection reach of current interferometers toward new astrophysical sources under realistic observational constraints.
