Reconstruction of Glitch-affected Gravitational Wave data using Artificial Neural Networks

Real-time Gravitational Wave data at LIGO encounters numerous glitches arising from known and unknown sources of noise. In cases where they occur in conjunction with an incoming Gravitational Wave (GW) signal, they can seriously hinder signal detection and its consequent analysis. Current techniques to handle such scenarios include applying a gated cut to the data segment which includes a glitch, and later carefully model the glitch to clean the data segment in question. Here, we explore the use of different machine learning regression models to reconstruct the glitch-affected regions of a data stream whenever the glitch appears over a GW signal. We compare Multi-layered Perceptron (MLP) based Neural Network towards this goal and present a proof of concept for a low-latency, glitch-independent method of cleaning and reconstructing glitch-affected data for a quick primary analysis of an incoming GW signal.