Improving the detection significance of gravitational wave transient searches with CNN models
Johann Fernandes, Archana Pai, Koustav Chandra · 2025
Read Improving the detection significance of gravitational wave transient searches with CNN models by Johann Fernandes, Archana Pai, Koustav Chandra (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-08332.txt.
Introduction
Johann Fernandes, Archana Pai, and Koustav Chandra work on spotting gravitational waves in a 2025 paper. Searches for these ripples must tell real signals from noise glitches. Standard tools are tuned to expected signal shapes, which can miss stranger events as detectors improve. The team tried a learning-based filter instead.
Their image-based neural network sorts glitches from true events and feeds that judgment into two existing searches. The searches then reach about a third farther in volume over time at a strict false alarm rate. Applied to real recorded data, the method raised the standing of several events, most strikingly the first very heavy black hole merger on record.
Catalog record
| CREATOR | Johann Fernandes, Archana Pai, Koustav Chandra |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 41 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-08332
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Improving the detection significance of gravitational wave transient searches with CNN models Johann Fernandes, Archana Pai, Koustav Chandra Abstract Gravitational wave (GW) transient searches rely on signal-noise discriminators to distinguish astrophysical signals from noise artefacts. These discriminators are typically tuned towards expected signal morphologies, which may limit their effectiveness as detector sensitivity improves and more complex signals, such as from core collapse supernovae or compact binary mergers featuring precession, higher-order harmonics, or eccentricity, become detectable. In this work, we use a Convolutional Neural Network-based approach to classify noise transients from astrophysical transients, aiming to enhance the sensitivity of existing searches. We evaluate our method on two matched filter based searches, PyCBC-IMBH and PyCBC-HM tuned for imbh ( imbh ) binary systems. Our approach improves the sensitive volume-time reach of these searches by approximately 30% at a false alarm rate of once per 100 years. Finally, we apply our method to the first four chunks of the first half of the third observation run and demonstrate a marked improvement in significance. In particular, we significantly improve the first IMBH binary gw ( gw ) event GW190521 with an IFAR exceeding 42000 yr. I Introduction The lvk ( lvk ) has observed approximately 100 gw transient events in the first three observational runs, with most of them being consistent with merging quasi-spherical compact objects ( Aasi et al. 2015 ; Acernese et al. 2015 ; Akutsu et al. 2019 ; Abbott et al. 2019 ; Abbott et al. 2021a ; Abbott et al. 2023a ) . This number is projected to at least double during the ongoing fourth observing run (O4) of the advanced lvk detectors and increase by orders of magnitude with the advent of xg ( xg ) gw detectors, such as ce ( ce ) and et ( et ) ( Punturo et al. 2010 ; Reitze et al. 2019 ; Maggiore et al. 2020a ; Evans et al. 2021 ; Branchesi et al. 2023 ; Gupta et al. 2024 ; Abbott et al. 2016a ; Hild et al. 2011 ; Abernathy et al. 2011 ; Abbott et al. 2017 ) . These detections, primarily bbh ( bbh ) mergers, have provided a unique opportunity to study bh , enabling precision tests of gr ( gr ) in the high-curvature, strong-field regime and offering insights into the astrophysical origins and nature of bh populations ( Abbott et al. 2021b ; Abbott et al. 2023b ; Abbott et al. 2024 ) . However, detecting and inferring the properties of these events becomes challenging, especially when the signal is short-lived. These short-lived signals can be either intrinsic, like supernova explosions, or a result of the limited bandwidth of the detector. In both cases, detecting such signals poses significant challenges due to the noise transients inherent in interferometric detectors. These non-Gaussian noise transients, also called glitches, are of terrestrial origin Davis et al. 2021 and are classified based on their morphology and/or source. Common categories include blips, scattered light, koi fish, tomtes, etc. ( Cabero et al. 2019 ; Zevin et al. 2017 ) . Mitigating these noise transients and improving the detection sensitivity of searches is an active area of research. As an example, consider the case of cbc ( cbc ) systems with heavier masses, especially when one of the component masses or the final bh is an imbh . The gw signal from such a system falls within the lower frequency range of the advanced lvk detectors. This results in a short-lived gw transient predominantly characterised by the merger and ringdown phases, with only a tiny contribution from the inspiral phase. Most notably, the first confident imbh binary event GW190521 ( Abbott et al. 2020 ) observed during the third observation run of lvk detectors was of this type, being barely observable within the detector bandwidth. As a result, GW190521 has been the subject of multiple interpretations, the most credible being a quasi-spherical imbh binary event. However, other analyses have also suggested possibilities that range from a hyperbolic encounter to the result of a head-on collision between exotic compact objects ( Fishbach and Holz 2020 ; Nitz and Capano 2021 ; Romero-Shaw et al. 2020 ; Gayathri et al. 2022 ; Calderón Bustillo et al. 2021 ; Estellés et al. 2022 ; Olsen et al. 2021 ; Calderon Bustillo et al. 2023 ; Chandra et al. 2024 ) . Consequently, template-based searches like PyCBC-broad, which use quasi-circular quadrupolar bbh waveforms, failed to identify this event with sufficient significance (see supplement of ( Abbott et al. 2020 ) ). Additionally, the PyCBC-broad search suffered from the “look-elsewhere effect”, as it spanned a broad parameter space of cbc systems ( Chandra et al. 2020 ) . As a result, its empirically derived background included noise triggers resembling GW190521. To overcome this challenge, the PyCBC-IMBH search was developed. It was specifically designed to target quasi-circular bbh signals with a detector-frame total mass M_{T}(1+z)/M\odot>100 ( Chandra et al. 2021a ; Chandra et al. 2021b ) , incorporating strict vetoes to suppress glitches. Meanwhile, the PyCBC-HM search ( Chandra et al. 2022 ) was tailored for mass-asymmetric edge-on intermediate-mass binary black hole (IMBH) systems. PyCBC-HM was the first matched-filter based approach for asymmetric IMBH binaries by incorporating higher-order modes into its quasi-circular template bank. To summarise, template-based searches often devise different signal-noise discriminators and incorporate them in the test statistics to downrank the noise triggers and improve the search’s signal-noise discriminatory ability ( Allen 2005a ; Nitz 2018 ; Messick et al. 2017 ) This requires dedicated efforts to identify new signal-noise discriminators that are independent of the existing ones. Weakly modelled searches, such as cwb ( cwb ), on the other hand, utilise the excess coherent power across a network of detectors, to detect signals. They make minimal assumptions about the signal morphology in the time-frequency plane. As with templated searches, they use a variety of signal-noise discriminators during the post-processing stage to compute the background and thus, the final significance of the gw triggers ( Klimenko et al. 2016 ) . As detector sensitivity improves, we anticipate observing more gw events that are currently very rare or remain unobserved ( Hall 2022 ; Maggiore et al. 2020b ) . However, we may also see additional glitches that could potentially display complex morphologies due to the larger frequency bandwidth of these detectors. Without effective signal-noise discriminators, we risk missing these exceptional events amidst the noise. It is crucial to improve searches as rare events can provide insights into the astrophysical properties of sources. For instance, imbh binaries can probe black hole spectroscopy via the remnant ( Ghosh et al. 2021 ; Ota and Chirenti 2022 ) , whereas kick velocities can inform us about the astrophysical formation channel of massive bh s ( Mahapatra et al. 2021 ; Fishbach et al. 2017 ; Gerosa and Berti 2017 ) . These can have a direct implication on astrophysics and bh physics. Over the past decade, several ml ( ml )-based gw algorithms, particularly those based on deep learning, have proven effective in rapid signal detection, parameter estimation, and glitch classification, among many other applications. For instance, ( Marx et al. 2025 ) utilise a