DeepClean: Machine Learning-Assisted Noise Regression in Gravitational Wave Detectors
ORAL
Abstract
The sensitivities of Gravitational wave (GW) detectors such as advanced LIGO, advanced Virgo, and KAGRA are often limited by instrumental and environmental effects. The noise from these sources couples non-linearly to the GW strain and goes beyond the capacities of the conventional filtering methods. In recent years, Machine Learning algorithms have been proven capable of removing such non-linear noise couplings. DeepClean is a convolutional neural network algorithm for subtracting non-linear and non-stationary noise from GW strain. To estimate the noise contamination, DeepClean uses the auxiliary witness sensors that independently record the instrumental and environmental random processes which cause the contamination. This work presents the results from a mock data challenge demonstrating DeepClean as a low-latency pipeline for noise regression in LIGO data. We benchmark the performances in terms of latency, signal-to-noise ratio, and astrophysical parameter estimation.
–
Presenters
Muhammed S Cholayil
LIGO, University of Minnesota
Authors
Muhammed S Cholayil
LIGO, University of Minnesota
Alec Gunny
Massachusetts Institute of Technology, LIGO Lab, MIT
Chia-Jui Chou
National Yang Ming Chiao Tung University, Taiwan
Li-Cheng Yang
National Yang Ming Chiao Tung University, Taiwan
Michael W Coughlin
University of Minnesota
William Benoit
University of Minnesota
Dylan S Rankin
Massachusetts Institute of Technology, University of Pennsylvania, MIT
Ethan J Marx
Massachusetts Institute of Technology
Deep Chatterjee
Massachusetts Institute of Technology, MIT
Erik Katsavounidis
Massachusetts Institute of Technology, MIT, LIGO Lab, MIT
