Detection of Gravitational Wave signals using Machine Learning methods and Generative Pre-trained Transformers

Abstract

We use Machine Learning methods based on Convolutional Neural Networks to search for gravitational waves signals above the background noise distribution for a data set of simulated gravitational waves and real noise signals from three detectors (LIGO Hanford, LIGO Livingston, and Virgo). A training data set is used to train the ML method to classify data streams in two groups: gravitational wave plus noise (label 1) or only noise (label 0). Later, the method predicts if data streams from a testing data set belong to one or an other category. To generate the code that implements the CNN algorithm we use Generative Pre-trained Transformers, specifically ChatGPT based on GPT-3 and compare them to a human-made CNN. The ML methods are capable to detect gravitational waves if we give ChatGPT freedom to create a CNN without specifying the parameters or the architecture, but are not satisfactory if we try to direct ChatGPT to a specific type of code.