Advancing Glitch Classification in Gravity Spy - Multi-view Fusion with Attention-based Machine Learning for Advanced LIGO’s Fourth Observing Run

Aug 5, 2026

Upload by Violet

Ground-based gravitational-wave observatories such as LIGO enable breakthrough observations of spacetime ripples, yet their extreme sensitivity also captures non-Gaussian noise transients known as glitches that can mimic or mask true signals. Gravity Spy addresses this by combining community expertise with machine learning to classify glitches in LIGO data. While an integrated classifier has operated since 2017, limitations emerged in LIGO’s ongoing fourth observing run (O4) due to simplistic architecture and weak multi-time-window generalization. This work proposes an advanced O4 classifier using prior-run data, evaluating multi-view fusion strategies, applying label smoothing for noisy labels, and improving interpretability via attention-based weighting. Results show improved performance to support advanced glitch classification for continued gravitational-wave research.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages29

Reading Time1 h 13 min

AccessFree to download

Introduction

Background: gravitational waves and detector noise

Gravity Spy and machine-learning-based glitch classification

Motivation: limits of the existing O4 classifier architecture

Proposed approach

Multi-time-window multi-view fusion strategies

Label smoothing for noisy labels

Attention-based interpretability through learned weights

arXiv :2401 . 12913v2 [gr-qc] 14 Aug 2025 Advancing Glitch Classification in Gravity Spy: Multi-view Fusion with Attention-based Machine Learning for Advanced LIGO’s Fourth Observing Run Yunan Wu 1 , Michael Zevin2 ,3 , Christopher P. L. Berry4 , Kevin Crowston5 , Carsten Østerlund5 , Zoheyr Doctor3 ,6 , Sharan Banagiri3 , Corey B. Jackson7 , Vicky Kalogera3 ,6 , Aggelos K. Katsaggelos 1 ,3 1 The Department of Electrical Computer Engineering, Northwestern University, 2145 Sheridan Road, Evanston, 60208, IL, USA 2 The Adler Planetarium, 1300 South DuSable Lake Shore Drive, Chicago, 60605, IL, USA 3 Center for Interdisciplinary Exploration and Research in Astrophysics (CIERA), Northwestern University, 1800 Sherman Ave, Evanston, 60201, IL, USA 4 SUPA, School of Physics and Astronomy, University of Glasgow, Kelvin Building, University Ave, Glasgow, 8QQ, G12, UK 5 School of Information Studies, Syracuse University, Hinds Hall, Syracuse, 13210, NY, USA 6 Department of Physics and Astronomy, Northwestern University, 2145 Sheridan Road, Evanston, 60208, IL, USA 7 Information School, University of Wisconsin–Madison, 600 N Park Street, Madison, 53706, WI, USA E-mail: [christopher.berry.2@glasgow.ac.uk](christopher.berry.2@glasgow.ac.uk) Abstract. The first successful detection of gravitational waves by ground-based observatories, such as the Laser Interferometer Gravitational-Wave Observatory (LIGO), marked a breakthrough in our comprehension of the Universe. However, due to the unprecedented sensitivity required to make such observations, gravitational-wave detectors also capture disruptive noise sources called glitches, which can potentially be confused for or mask gravitational-wave signals. To address this problem, a community-science project, Gravity Spy, incorporates human insight and machine learning to classify glitches in LIGO data. The machine-learning classifier, integrated into the project since 2017, has evolved over time to accommodate increasing numbers of glitch classes. Despite its success, limitations have arisen in the ongoing LIGO fourth observing run (O4) due to the architecture’s simplicity, which led to poor generalization and inability to handle multi-time window inputs effectively. We propose an advanced classifier for O4 glitches. Using data from previous observing runs, we evaluate different fusion strategies for multi-time window inputs, using label smoothing to counter noisy labels, and enhancing interpretability through attention modulegenerated weights. Our new O4 classifier shows improved performance, and will enhance glitch classification, aiding in the ongoing exploration of gravitational-wave phenomena. Multi-view Fusion with Attention-based Machine Learning for O4 2 Submitted to: Class. Quantum Grav. 1. Introduction The first discovery of gravitational waves, a pivotal element in Einstein’s theory of general relativity [1], has opened up an entirely new window in the cosmos. The Laser Interferometer Gravitational-Wave Observatory (LIGO) [2] achieved groundbreaking success in detecting these ripples in spacetime for the first time in September 2015 [3] . Since then, the gravitational-wave detector network has been expanded with the addition of Virgo [4], which joined observations in August 2017 [5, 6], and KAGRA [7], which initiated its observing run in April 2020 [8]; this network has collected a large catalog of gravitational-wave observations [9] . However, collecting these observations demands exceptionally sensitive and intricate detectors in order to be able to measure minuscule fluctuations in spacetime [10] . This heightened sensitivity, in turn, leads to the detection of diverse sources of noise, with the potential to obscure or mimic authentic gravitational-wave signals [9, 11–13] . Bursts of non-Gaussian noise caused by environmental or instrumental factors, known as glitches, are particularly disruptive to measuring gravitational waves. The origins of many gli

FAQs

Why is glitch classification necessary for LIGO gravitational-wave searches?

Glitches are disruptive non-Gaussian noise transients that can obscure or mimic gravitational-wave signals. Classifying and removing them improves detector-data quality and analysis reliability.

What limitations affected the Gravity Spy machine-learning classifier in LIGO’s O4?

The ongoing O4 revealed weaknesses from the classifier’s relatively simple architecture, including poor generalization and difficulty handling multi-time-window inputs effectively.

What key methods does the proposed O4 classifier introduce?

It evaluates fusion strategies for multi-time-window inputs, uses label smoothing to mitigate noisy labels, and adds attention-based modules to produce interpretable weighting for the model’s decisions.

Aug 5, 2026

Upload by Violet