Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Pre-trained Audio Transformer as a Foundational AI Tool for Gravitational Waves

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A pretrained audio transformer, lightly fine-tuned, detects gravitational waves and classifies glitches.

desk verdict A useful demonstration that a compact audio transformer can be adapted for GW detection, but the transfer claim lacks an ablation and the O3b FAR numbers are extrapolated beyond what the data supports. read the letter →

arxiv 2412.20789 v2 pith:EP4ABUFQ submitted 2024-12-30 gr-qc astro-ph.HE

Chayan Chatterjee , Abigail Petulante , Karan Jani , Jesse Spencer-Smith , Yang Hu , Roy Lau , Haowei Fu , Trang Hoang
show 2 more authors
Stephen Chong Zhao Suyash Deshmukh
This is my paper · ORCID
classification gr-qcastro-ph.HE
keywords gravitationalwavesfoundationmodelsWhispertransformerQ-transformglitchclassificationparameter-efficientfine-tuningMLGWSC-1
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

GW-Whisper adapts Whisper, a transformer pretrained on hundreds of thousands of hours of human speech, to gravitational-wave data by replacing its audio front end with Q-transform spectrograms and fine-tuning only 0.5% of the model's parameters. The paper claims this is the first cross-domain repurposing of an open-source audio transformer for gravitational-wave research, and that the adapted model reaches a sensitive distance of about 900 Mpc at one false alarm per month on both MLGWSC-1 datasets 3 and 4. On real O3b data it recovers 18 of 31 confident binary black-hole mergers reported in GWTC-3, with 16 of those below one false alarm per month. For glitch classification, the original log-mel front end outperforms the GW-specific variant, correctly labeling true gravitational-wave signals 94% of the time in the generic test and 95% in the high-mass test. The case matters because if a speech model transfers this well, foundation models could provide scalable, low-cost analysis tools as detection rates climb.

What carries the argument

The load-bearing mechanism is the pretrained Whisper encoder itself, a 24-layer transformer originally trained on 80-band log-mel spectrograms of speech, combined with three lightweight adaptations. A Q-Adapter, a small convolutional stack, reshapes per-detector Q-transform spectrograms into the encoder's expected $80 \times 3000$ input and applies learned per-detector scale and shift (FiLM) parameters. DoRA (Weight-Decomposed Low-Rank Adaptation) updates only the query, key, value, and output projection matrices of the multi-head attention layers, training just 196,608 of the model's 39 million parameters. A self-supervised contrastive pre-training step on noise-invariant views of injected waveforms initializes the Q-Adapter and encoder before supervised fine-tuning. For the O3b false-alarm claims, an exponential fit to the upper 5% of the one-month background ranking distribution extrapolates false-alarm rates into the zero-count regime.

What would settle it

Compute the false-alarm rate for GW-Whisper's O3b triggers using a much longer background, for example all science-quality O3 data or an injection-based noise simulation, and count how often noise alone produces ranking statistics as high as those of the 16 claimed events; if the observed rate exceeds the extrapolated sub-$10^{-20}$-per-month values, the zero-count extrapolation fails.

Watch

Extended reading notes

Core claim

The central claim is that the learned representations of a pretrained audio transformer carry enough general time-frequency structure that they can be redirected to gravitational-wave analysis with minimal retraining. Concretely, GW-Whisper feeds per-detector Q-transform spectrograms through a lightweight Q-Adapter with feature-wise linear modulation, then through the frozen Whisper encoder whose attention projections are updated with low-rank DoRA adapters, and finally through a small MLP head. On the MLGWSC-1 benchmark the adapted model achieves about 900 Mpc sensitive distance at one false alarm per month on datasets 3 and 4, outperforming two of the machine-learning pipelines in the comparison while falling below the matched-filter and unmodeled searches. Applied to O3b data, the paper reports recovery of 18 of 31 confident BBH events, 16 below one false alarm per month, with several very loud events assigned extrapolated false-alarm rates below $10^{-20}$ per month. The paper also finds that keeping Whisper's original log-mel spectrogram front end is better for glitch classification than the GW-specialized Q-scan front end, which it interprets as evidence that audio-domain priors transfer to transient noise artifacts.

Load-bearing premise

The O3b recovery claim rests on the assumption that an exponential curve fitted to the upper 5% of one month of background triggers accurately models how often very high ranking statistics occur, so false-alarm rates as low as $10^{-20}$ per month can be assigned to events never seen in the background.

Editorial extensions

If this is right

  • A single pretrained audio encoder can serve both signal detection and glitch classification after fine-tuning only 0.5% of its parameters, lowering the computational barrier for real-time gravitational-wave searches.
  • At one false alarm per month, the adapted model reaches about 900 Mpc sensitive distance on realistic O3a noise, showing that speech-pretrained representations compete with dedicated machine-learning search pipelines.
  • Recovering 18 of 31 confident O3b BBH events, 16 below one false alarm per month, suggests template-free searches built on audio foundation models can identify catalog mergers without matched filtering.
  • Keeping the original log-mel front end for glitch classification outperforms the GW-specific Q-scan front end, indicating that speech-trained time-frequency features transfer better to transient noise than chirp-optimized features.
  • Parameter-efficient fine-tuning at the 0.5% level makes the approach scalable to the higher detection rates expected in future observing runs.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The exponential-tail extrapolation that assigns sub-$10^{-20}$ per month false-alarm rates to O3b candidates is the paper's most fragile step; a longer background or injection-based false-alarm estimate would test whether those rates hold.
  • Because Whisper-tiny is the smallest model in the Whisper family, larger pretrained audio transformers may yield higher sensitivity, though at greater computational cost; the paper does not test this.
  • The Q-Adapter plus FiLM recipe is generic: the same architecture could be pointed at other time-frequency data, such as continuous-wave searches or detector-characterization tasks, without redesigning the encoder.
  • A hybrid input that combines spectrogram and raw time-series features, which the paper names as future work, would likely close the remaining sensitivity gap to matched filtering at low signal-to-noise ratios.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper introduces GW-Whisper, an adaptation of OpenAI's Whisper-tiny audio transformer for gravitational-wave (GW) detection and glitch classification. The authors replace Whisper's log-mel front end with a per-detector Q-transform and a learned Q-Adapter, add FiLM conditioning, perform contrastive pre-training on injected waveforms, and fine-tune with DoRA adapters plus an MLP head. The claimed results are a sensitive distance of about 900 Mpc at a false alarm rate of one per month on MLGWSC-1 datasets 3 and 4, recovery of 18 of 31 confident O3b BBH events from GWTC-3, and high single-class accuracies in generic and high-mass glitch classification tasks. The central claim is that this is the first cross-domain application of an open-source audio transformer to GW research, demonstrating that a model pre-trained on human speech can be repurposed for GW analysis.

Significance. If established, the result would be a useful demonstration that a large, openly available audio foundation model can be adapted with lightweight fine-tuning to GW detection and glitch classification, with potential computational savings relative to training from scratch. The paper has concrete strengths: it evaluates on the standardized MLGWSC-1 benchmark against several published pipelines, it applies the model to real O3b data following an existing search setup, it reports parameter-efficient fine-tuning with only 0.5% of parameters trained, and it states code and data availability. The main significance hinges on the transfer claim, which is currently not isolated by any controlled comparison against a randomly initialized encoder, and on the O3b false alarm rates, which rest on an extrapolation that is not validated. With those points addressed, the contribution would be a solid empirical study of foundation-model transfer in GW data analysis.

major comments (3)
  1. [Abstract and Fig. 1 architecture] The central claim that speech pre-training transfers to GW tasks is not established because the paper never compares the pre-trained Whisper encoder with a randomly initialized encoder of the same architecture. The final model adds a GW-specific Q-Adapter, FiLM conditioning, contrastive pre-training on injected BBH waveforms, DoRA adapters, and a new MLP head, so the measured benchmark and O3b performance is compatible with the alternative hypothesis that these components, rather than the speech pre-training, carry the performance. A controlled ablation that trains a randomly initialized copy of the same encoder with the same data, adapters, and training budget is needed before the title/abstract claim of cross-domain transfer can be accepted.
  2. [Sec. 1.1, Eq. (4), and Fig. 3] The O3b false alarm rates below one per month are load-bearing for the claim that 16 of the 18 recovered events are significant, but they rest on an exponential extrapolation of a one-month O3a background over roughly 20 orders of magnitude into the zero-count regime. The fit to the upper 5% of the background distribution is not validated against an independent background, an extreme-value model, or any test of O3a-to-O3b noise stationarity. The authors should either provide validation of the tail behavior and uncertainty estimates for the extrapolated FARs, or downgrade the O3b significance claims to statements based on empirical count-based upper limits that do not require the exponential assumption.
  3. [Sec. 1.2 and Fig. 4] The glitch classification results are reported as single accuracy numbers from confusion matrices, but the manuscript does not state the dataset sizes, class balance, train/validation/test split, number of independent runs, or error bars. Without explicit train/test separation and statistical uncertainty, the reported accuracies such as 94% for generic GW and 95% for high-mass GW cannot be distinguished from overfitting. The authors should report the exact split procedure, per-class sample counts, and standard errors or confidence intervals, ideally with repeated-seed results.
minor comments (6)
  1. [Fig. 2 caption] The caption contains a typo: 'correspon' should be 'correspond'.
  2. [Discussion, final paragraph] There is a typo in 'GW-Whiaper' that should read 'GW-Whisper'.
  3. [Fig. 1 caption and Sec. 3] The caption describes the Whisper encoder as 'frozen' while the text and Methods state that DoRA adapters fine-tune the multi-head attention output-projection weights; this should be reconciled to avoid confusion about which parts are frozen and which are trainable.
  4. [Introduction and Sec. 3] The MLP head is described as 512 → 256 → 128 → 64 → 2 units in the introduction but as three hidden layers with 1024, 512, and 256 neurons in the Methods; the actual architecture should be stated consistently.
  5. [Sec. 1.1, Eq. (4)] The description of the exponential fit would be clearer if it stated explicitly that ln(FAR) is fit as a linear function of the ranking statistic and reported the fitted coefficients a and b with their uncertainties.
  6. [Sec. 3, contrastive pre-training] The phrase 'negative pairs (two pure-noise clips)' is confusing because pure-noise clips are not used as a separate negative-pair objective in the InfoNCE loss; the loss treats other batch elements as negatives. This should be reworded.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark and O3b results are external empirical evaluations, and the fitted FAR tail is a calibration extrapolation rather than a constructed prediction.

full rationale

The paper's quantitative claims are evaluated against external benchmarks and catalogs rather than derived from its own fitted constants. The sensitive distance on MLGWSC-1 datasets 3 and 4 is computed from injected signals in a standardized external challenge, and the O3b recovery of 18 of 31 events is defined by cross-matching candidate triggers to the independent GWTC-3 catalog within 0.2 s. The only fitted quantity in the reporting path is the exponential false-alarm tail in Eq. (4), which is fit to the upper 5% of one month of O3a background and then extrapolated to assign FARs to zero-count O3b candidates. That is a calibration and extrapolation choice, not an input renamed as a prediction, and it does not enter the MLGWSC-1 benchmark results. The central transfer claim—that Whisper's speech pretraining contributes to gravitational-wave performance—is not established by an ablation against a randomly initialized encoder, but that is an evidentiary gap rather than circularity: no passage defines the pretraining benefit in terms of the measured outcome, and the paper does not rely on a self-citation chain or a uniqueness theorem to force its conclusion. The design includes Q-Adapter, FiLM, contrastive pretraining, and DoRA fine-tuning, so the possibility that those components carry the performance is a real alternative hypothesis, but it is not a reduction of the claim to its own inputs. References to Whisper, MLGWSC-1, GWTC-3, and external pipelines are independent sources, and no load-bearing step is justified solely by the authors' own prior work. The O3b FAR extrapolation is a legitimate correctness and robustness risk, not a circularity. Overall score 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claims rest on transferability of Whisper's pretrained features, the adequacy of Q-transform inputs, the representativeness of the MLGWSC-1 benchmark, and the extrapolated false alarm model. No new physical entities, forces, or particles are introduced; Q-Adapter, FiLM, and DoRA are architectural components rather than invented physical entities.

free parameters (5)
  • DoRA rank = 8
    Chosen low-rank dimension for the PEFT adapters; controls how much of the pretrained model is modified.
  • Contrastive temperature tau = not reported
    Hyperparameter in the InfoNCE loss used in contrastive pre-training; no numerical value is given in the paper.
  • Clustering window Delta-t = 0.2 s
    Chosen window for grouping first-level triggers into events; affects event matching and false alarm counting.
  • Exponential FAR tail coefficients a and b = not reported
    Least-squares fit to the upper 5% of the background ranking distribution, used to extrapolate false alarm rates in the zero-count regime (Eq. 4).
  • Evaluation window length and stride = 1 s / 0.1 s
    Slicing choices for evaluating continuous data; affect trigger density and clustering.
assumptions (5)
  • domain assumption Whisper's speech-pretrained encoder representations transfer to gravitational-wave time-frequency data after light adaptation.
    The entire approach assumes pretrained audio features are useful for GW inputs. This is tested empirically but not proven.
  • domain assumption Q-transform spectrograms resized to 80 by 3000 preserve sufficient signal information for detection.
    The paper notes fixed spectrogram inputs are suboptimal for low signal-to-noise ratios, so this assumption is only partially satisfied.
  • ad hoc to paper Exponential tail extrapolation of the background ranking statistic describes the zero-count false alarm regime.
    Equation 4 extends empirical false alarm rates beyond observed data and is used to assign very small FARs to O3b candidates.
  • domain assumption MLGWSC-1 datasets 3 and 4 are representative and independent of the training injection distribution.
    The benchmark is standard, but the paper does not describe train/test splits or confirm that no overlap exists.
  • domain assumption IMRPhenomXPHM waveforms adequately model training and evaluation signals.
    The waveform model is used for injections in training and validation and is standard in the field.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pre-trained Audio Transformer as a Foundational AI Tool for Gravitational Waves." pith.science (2026). https://pith.science/paper/EP4ABUFQ

@misc{pith2026241220789,
  author       = {Pith},
  title        = {Pith review of: Pre-trained Audio Transformer as a Foundational AI Tool for Gravitational Waves},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EP4ABUFQ}},
  note         = {Machine review of arXiv:2412.20789}
}
read the original abstract

As gravitational wave detectors become more advanced and sensitive, the number of signals recorded by Advanced LIGO and Virgo from merging compact objects is expected to rise dramatically. This surge in detection rates necessitates the development of adaptable, scalable, and efficient tools capable of addressing a wide range of tasks in gravitational wave astronomy. Foundational AI models present a transformative opportunity in this context by providing a unified framework that can be fine tuned for diverse applications while leveraging the power of large scale pre training. In this work, we explore how advanced transformer models, specifically Whisper by OpenAI, can be adapted as a foundational model for gravitational wave data analysis. By fine tuning the encoder model of Whisper, originally trained on extensive audio data, and combining it with neural networks for specialized tasks, we achieve reliable results in detecting astrophysical signals and classifying transient noise artifacts or glitches. This represents the first application of open source transformer models, pre trained on unrelated tasks, for gravitational wave research, demonstrating their potential to enable versatile and efficient data analysis in the era of rapidly increasing detection rates.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Transformers with Physics-Informed Encodings and Simulation-Based Inference for Robust Detection of Eccentric Binary Black Holes in Pulsar Timing Array Data

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Physics-informed Transformer encodings plus conditional normalizing flows yield sharper, better-calibrated posteriors for eccentric BBHs in white-noise PTA data than physics-agnostic SBI baselines.

Reference graph

Works this paper leans on

62 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [1]

    Abbott, B.P., Abbott, R., Abbott, T.D., Abernathy, M.R., Acernese, F., Ackley, K., Adams, C., Adams, T., Addesso, P., Adhikari, R.X., Adya, V.B., Affeldt, C., Agathos, M., Agatsuma, K., Aggarwal, N., Aguiar, O.D., Aiello, L.: Observation of gravitational waves from a binary black hole merger. Phys. Rev. Lett. 116, 061102 (2016) https://doi.org/10.1103/Phy...

    doi:10.1103/physrevlett.116.061102 2016

  2. [2]

    Classical and Quantum Gravity 32(7), 074001 (2015) https://doi.org/10.1088/0264-9381/ 32/7/074001

    Aasi, J., Abbott, B.P., Abbott, R., Abbott, T., Abernathy, M.R., Ackley, K., Adams, C., Adams, T., Addesso, P., Adhikari, R.X., Adya, V., Affeldt, C., Aggar- wal, N., Aguiar, O.D., Ain, A., Ajith, P., Alemic, A., Allen, B., Amariutei, D., Anderson, S.B., Anderson, W.G., Collaboration, L.S.: Advanced ligo. Classical and Quantum Gravity 32(7), 074001 (2015)...

    doi:10.1088/0264-9381/ 2015

  3. [3]

    Abbott, B.P., Abbott, R., Abbott, T.D., Abraham, S., Acernese, F., Ackley, K., Adams, C., Adhikari, R.X., Adya, V.B., Affeldt, C., Agathos, M., Agatsuma, K., Aggarwal, N., Aguiar, O.D., Aiello, L., Ain, A., Ajith, P., Allen, G., Allocca, A., Aloy, M.A., Altin, P.A., Amato, A., Ananyeva, A., Anderson, S.B., Anderson, W.G., Angelova, S.V., Antier, S., Apper...

    work page 2019

  4. [4]

    Abbott, R., Abbott, T.D., Abraham, S., Acernese, F., Ackley, K., Adams, A., Adams, C., Adhikari, R.X., Adya, V.B., Affeldt, C., Agathos, M., Agatsuma, K., Aggarwal, N., Aguiar, O.D., Aiello, L., Ain, A., Ajith, P., Akcay, S., Allen, G., Allocca, A., Altin, P.A., Amato, A., Anand, S., Ananyeva, A., Anderson, S.B., Anderson, W.G., Angelova, S.V.: Gwtc-2: Co...

    2021

  5. [5]

    Abbott, R., Abbott, T.D., Acernese, F., Ackley, K., Adams, C., Adhikari, N., Adhikari, R.X., Adya, V.B., Affeldt, C., Agarwal, D., Agathos, M., Agatsuma, K., Aggarwal, N., Aguiar, O.D., Aiello, L., Ain, A., Ajith, P., Albanesi, S., Allocca, A., Altin, P.A., Amato, A., Anand, C., Anand, S., Ananyeva, A., Anderson, S.B., Anderson, W.G., Andrade, T., Andres,...

    2024

  6. [6]

    Abbott, R., Abbott, T.D., Acernese, F., Ackley, K., Adams, C., Adhikari, N., Adhikari, R.X., Adya, V.B., Affeldt, C., Agarwal, D., Agathos, M., Agatsuma, K., Aggarwal, N., Aguiar, O.D., Aiello, L., Ain, A., Ajith, P., Akcay, S., Akutsu, T., Albanesi, S., Allocca, A., Altin, P.A., Amato, A., Anand, C., Anand, S., Ananyeva, A., Anderson, S.B., Anderson, W.G...

    work page 2023

  7. [7]

    Nature 551(7678), 85–88 (2017) https://doi.org/10.1038/nature24471

    Abbott, B.P., Abbott, R., Abbott, T.D., Acernese, F., Ackley, K., Adams, C., Adams, T., Addesso, P., Adhikari, R.X., Adya, V.B., Affeldt, C., Afrough, M., Agarwal, B., Agathos, M., Collaboration, T.L.S., Collaboration, T.V.: A gravitational-wave standard siren measurement of the hubble constant. Nature 551(7678), 85–88 (2017) https://doi.org/10.1038/nature24471

    doi:10.1038/nature24471 2017

  8. [8]

    Berti, E., Yagi, K., Yunes, N.: Extreme Gravity Tests with Gravitational Waves from Compact Binary Coalescences: (I) Inspiral-Merger. Gen. Rel. Grav. 50(4), 46 (2018) https://doi.org/10.1007/s10714-018-2362-8 arXiv:1801.03208 [gr-qc]

    arXiv 2018

Show all 62 references
  1. [9]

    Abbott, B.P., Abbott, R., Abbott, T.D., Acernese, F., Ackley, K., Adams, C., 20 Adams, T., Addesso, P., Adhikari, R.X., Adya, V.B., Affeldt, C., Agarwal, B., Agathos, M., Agatsuma, K., Aggarwal, N., Aguiar, O.D., Aiello, L., Ain, A., Ajith, P., Allen, B., Allen, G., Allocca, A...

    2019

  2. [10]

    Classical and Quantum Gravity 32(2), 024001 (2014) https://doi.org/10.1088/ 0264-9381/32/2/024001

    Acernese, F., Agathos, M., Agatsuma, K., Aisa, D., Allemandou, N., Allocca, A., Amarni, J., Astone, P., Balestri, G., Ballardin, G., Barone, F., Baronick, J.-P., Barsuglia, M., Basti, A., Basti, F., Bauer, T.S., Bavigadda, V., Bejger, M., Beker, M.G., Belczynski, C., Bersanett...

    2014

  3. [11]

    https://arxiv.org/abs/2411.05754

    Maggiore, M., Iacovelli, F., Belgacem, E., Mancarella, M., Muttoni, N.: Compar- ison of global networks of third-generation gravitational-wave detectors (2024). https://arxiv.org/abs/2411.05754

    2024

  4. [12]

    The Astrophysical Journal 941(2), 208 (2022) https://doi.org/10.3847/1538-4357/ ac9cd4

    Iacovelli, F., Mancarella, M., Foffa, S., Maggiore, M.: Forecasting the detection capabilities of third-generation gravitational-wave detectors using gwfast. The Astrophysical Journal 941(2), 208 (2022) https://doi.org/10.3847/1538-4357/ ac9cd4

    2022 doi

  5. [13]

    The Astrophysical Journal Letters 913(1), 5 (2021) https://doi.org/10.3847/ 2041-8213/abf8be

    Ng, K.K.Y., Vitale, S., Farr, W.M., Rodriguez, C.L.: Probing multiple popu- lations of compact binaries with third-generation gravitational-wave detectors. The Astrophysical Journal Letters 913(1), 5 (2021) https://doi.org/10.3847/ 2041-8213/abf8be

    2021

  6. [14]

    Finn, L.S.: Aperture synthesis for gravitational-wave data analysis: Deterministic sources. Phys. Rev. D 63, 102001 (2001) https://doi.org/10.1103/PhysRevD.63. 102001

    2001 doi

  7. [15]

    The Astrophysical Journal Supplement Series 241(2), 27 (2019) https://doi.org/10.3847/1538-4365/ab06fc

    Ashton, G., H¨ ubner, M., Lasky, P.D., Talbot, C., Ackley, K., Biscoveanu, S., Chu, Q., Divakarla, A., Easter, P.J., Goncharov, B., Vivanco, F.H., Harms, J., Lower, M.E., Meadors, G.D., Melchor, D., Payne, E., Pitkin, M.D., Powell, J., Sarin, N., Smith, R.J.E., Thrane, E.: Bil...

    2019 doi

  8. [16]

    Veitch, J., Raymond, V., Farr, B., Farr, W., Graff, P., Vitale, S., Aylott, B., Blackburn, K., Christensen, N., Coughlin, M., Del Pozzo, W., Feroz, F., Gair, J., Haster, C.-J., Kalogera, V., Littenberg, T., Mandel, I., O’Shaughnessy, R., Pitkin, M., Rodriguez, C., R¨ over, C.,...

    2015

  9. [17]

    MIT Press, ??? (2016)

    Goodfellow, I., Bengio, Y., Courville, A.: Deep Learning. MIT Press, ??? (2016). http://www.deeplearningbook.org

    2016

  10. [18]

    Physics Letters B 778, 64–70 (2018) https://doi.org/10.1016/j.physletb.2017.12.053

    George, D., Huerta, E.A.: Deep learning for real-time gravitational wave detection and parameter estimation: Results with advanced ligo data. Physics Letters B 778, 64–70 (2018) https://doi.org/10.1016/j.physletb.2017.12.053

    2018 doi

  11. [19]

    Sch¨ afer, M.B., Zelenka, O.c.v., Nitz, A.H., Wang, H., Wu, S., Guo, Z.-K., Cao, Z., Ren, Z., Nousi, P., Stergioulas, N., Iosif, P., Koloniari, A.E., Tefas, A., Passalis, N., Salemi, F., Vedovato, G., Klimenko, S., Mishra, T., Br¨ ugmann, B., Cuoco, E., Huerta, E.A., Messenger...

    2023 doi

  12. [20]

    Gabbard, H., Williams, M., Hayes, F., Messenger, C.: Matching matched filter- ing with deep networks for gravitational-wave astronomy. Phys. Rev. Lett. 120, 141103 (2018) https://doi.org/10.1103/PhysRevLett.120.141103

    2018 doi

  13. [21]

    https: //arxiv.org/abs/2403.18661

    Marx, E., Benoit, W., Gunny, A., Omer, R., Chatterjee, D., Venterea, R.C., Wills, L., Saleem, M., Moreno, E., Raikman, R., Govorkova, E., Rankin, D., Coughlin, M.W., Harris, P., Katsavounidis, E.: A machine-learning pipeline for real-time detection of gravitational waves from ...

    2024 arXiv

  14. [22]

    Physics Letters B 840, 137850 (2023) https://doi.org/10.1016/j.physletb.2023.137850

    Qiu, R., Krastev, P.G., Gill, K., Berger, E.: Deep learning detection and clas- sification of gravitational waves from neutron star-black hole mergers. Physics Letters B 840, 137850 (2023) https://doi.org/10.1016/j.physletb.2023.137850

    2023

  15. [23]

    Physics Letters B 803, 135330 (2020) https: //doi.org/10.1016/j.physletb.2020.135330

    Krastev, P.G.: Real-time detection of gravitational waves from binary neutron stars using artificial neural networks. Physics Letters B 803, 135330 (2020) https: //doi.org/10.1016/j.physletb.2020.135330

    2020

  16. [24]

    Skliris, V., Norman, M.R.K., Sutton, P.J.: Real-Time Detection of Unmodelled Gravitational-Wave Transients Using Convolutional Neural Networks (2022)

    2022

  17. [25]

    Dax, M., Green, S.R., Gair, J., P¨ urrer, M., Wildberger, J., Macke, J.H., Buo- nanno, A., Sch¨ olkopf, B.: Neural importance sampling for rapid and reliable gravitational-wave inference. Phys. Rev. Lett. 130, 171403 (2023) https://doi. org/10.1103/PhysRevLett.130.171403

    2023 doi

  18. [26]

    Nature Physics18(1), 112–117 (2022) https://doi

    Gabbard, H., Messenger, C., Heng, I.S., Tonolini, F., Murray-Smith, R.: Bayesian parameter estimation using conditional variational autoencoders for gravitational-wave astronomy. Nature Physics18(1), 112–117 (2022) https://doi. org/10.1038/s41567-021-01425-7 22

    2022 doi

  19. [27]

    Williams, M.J., Veitch, J., Messenger, C.: Nested sampling with normalizing flows for gravitational-wave inference. Phys. Rev. D 103, 103006 (2021) https://doi. org/10.1103/PhysRevD.103.103006

    2021 doi

  20. [28]

    Chua, A.J.K., Vallisneri, M.: Learning bayesian posteriors with neural networks for gravitational-wave inference. Phys. Rev. Lett.124, 041102 (2020) https://doi. org/10.1103/PhysRevLett.124.041102

    2020 doi

  21. [29]

    Machine Learning: Science and Technology 3(1), 015007 (2021) https://doi.org/10.1088/2632-2153/ ac3843

    Shen, H., Huerta, E.A., O’Shea, E., Kumar, P., Zhao, Z.: Statistically-informed deep learning for gravitational wave parameter estimation. Machine Learning: Science and Technology 3(1), 015007 (2021) https://doi.org/10.1088/2632-2153/ ac3843

    2021 doi

  22. [30]

    Classical and Quantum Gravity 33(13), 134001 (2016) https://doi.org/10.1088/0264-9381/33/13/134001

    Abbott, B.P., Abbott, R., Abbott, T.D., Abernathy, M.R., Acernese, F., Ackley, K., Adamo, M., Adams, C., Adams, T., Addesso, P., Adhikari, R.X., Adya, V.B., Affeldt, C., Agathos, M., Agatsuma, K., Aggarwal, N., Aguiar, O.D., Aiello, L., Ain, A., Ajith, P., Allen, B., Allocca, ...

    2016

  23. [31]

    https://arxiv.org/abs/2409.02831

    Soni, S., Berger, B.K., Davis, D., Renzo, F.D., Effler, A., Ferreira, T.A., Glanzer, J., Goetz, E., Gonz´ alez, G., Helmling-Cornell, A., Hughey, B., Huxford, R., Man- nix, B., Mo, G., Nandi, D., Neunzert, A., Nichols, S., Pham, K., Renzini, A.I., Schofield, R.M.S., Stuver, A....

    2024

  24. [32]

    Classical and Quantum Gravity 38(13), 135014 (2021) https://doi.org/10.1088/1361-6382/abfd85

    Davis, D., Areeda, J.S., Berger, B.K., Bruntz, R., Effler, A., Essick, R.C., Fisher, R.P., Godwin, P., Goetz, E., Helmling-Cornell, A.F., Hughey, B., Katsavouni- dis, E., Lundgren, A.P., Macleod, D.M., M´ arka, Z., Massinger, T.J., Matas, A., McIver, J., Mo, G., Mogushi, K., N...

    2021

  25. [33]

    Classical and Quantum Gravity 34(6), 064003 (2017) 28 https://doi.org/10.1088/1361-6382/aa5cea

    Zevin, M., Coughlin, S., Bahaadini, S., Besler, E., Rohani, N., Allen, S., Cabero, M., Crowston, K., Katsaggelos, A.K., Larson, S.L., Lee, T.K., Lintott, C., Litten- berg, T.B., Lundgren, A., Østerlund, C., Smith, J.R., Trouille, L., Kalogera, V.: Gravity spy: integrating adva...

    2017 doi

  26. [34]

    Classical and Quantum Gravity38(19), 195016 (2021) https://doi.org/10.1088/1361-6382/ac1ccb

    Soni, S., Berry, C.P.L., Coughlin, S.B., Harandi, M., Jackson, C.B., Crowston, K., Østerlund, C., Patane, O., Katsaggelos, A.K., Trouille, L., Baranowski, V.- G., Domainko, W.F., Kaminski, K., Rodriguez, M.A.L., Marciniak, U., Nauta, P., Niklasch, G., Rote, R.R., T´ egl´ as, B...

    2021

  27. [35]

    Advances in Neural Information Processing Systems 33, 1877–1901 (2020)

    Brown, T.B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al.: Language models are few-shot learners. Advances in Neural Information Processing Systems 33, 1877–1901 (2020)

    2020

  28. [36]

    International Conference on Machine Learning (ICML) (2021)

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. International Conference on Machine Learning (ICML) (2021)

    2021

  29. [37]

    International Conference on Machine Learning (ICML) (2021)

    Ramesh, A., Pavlov, M., Goh, G., Gray, S., Voss, C., Radford, A., Chen, M., Sutskever, I.: Zero-shot text-to-image generation. International Conference on Machine Learning (ICML) (2021)

    2021

  30. [38]

    https:// arxiv.org/abs/2212.04356

    Radford, A., Kim, J.W., Xu, T., Brockman, G., McLeavey, C., Sutskever, I.: Robust Speech Recognition via Large-Scale Weak Supervision (2022). https:// arxiv.org/abs/2212.04356

    2022 arXiv

  31. [39]

    CoRR abs/1706.03762 (2017) 1706.03762

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. CoRR abs/1706.03762 (2017) 1706.03762

    2017 arXiv

  32. [40]

    https://api.semanticscholar.org/CorpusID:5960831

    Rabiner, L.R., Schafer, R.W.: Theory and application of digital speech processing by. https://api.semanticscholar.org/CorpusID:5960831

  33. [41]

    Proceedings of the IEEE 86(11), 2278–2324 (1998) https: //doi.org/10.1109/5.726791

    Lecun, Y., Bottou, L., Bengio, Y., Haffner, P.: Gradient-based learning applied to document recognition. Proceedings of the IEEE 86(11), 2278–2324 (1998) https: //doi.org/10.1109/5.726791

    1998 doi

  34. [42]

    https://arxiv.org/abs/1709

    Perez, E., Strub, F., Vries, H., Dumoulin, V., Courville, A.: FiLM: Visual Rea- soning with a General Conditioning Layer (2017). https://arxiv.org/abs/1709. 07871

    2017

  35. [43]

    https://arxiv.org/abs/1910.03771

    Wolf, T., Debut, L., Sanh, V., Chaumond, J., Delangue, C., Moi, A., Cistac, P., Rault, T., Louf, R., Funtowicz, M., Davison, J., Shleifer, S., Platen, P., Ma, C., Jernite, Y., Plu, J., Xu, C., Scao, T.L., Gugger, S., Drame, M., Lhoest, Q., Rush, A.M.: HuggingFace’s Transformer...

    2020 arXiv

  36. [44]

    WSEAS Trans

    Popescu, M.-C., Balas, V.E., Perescu-Popescu, L., Mastorakis, N.: Multilayer perceptron and neural networks. WSEAS Trans. Cir. and Sys. 8(7), 579–588 (2009)

    2009

  37. [45]

    https://arxiv.org/abs/2312.12148

    Xu, L., Xie, H., Qin, S.-Z.J., Tao, X., Wang, F.L.: Parameter-Efficient Fine- Tuning Methods for Pretrained Language Models: A Critical Review and Assessment (2023). https://arxiv.org/abs/2312.12148

    2023 arXiv

  38. [46]

    https: //arxiv.org/abs/2402.09353

    Liu, S.-Y., Wang, C.-Y., Yin, H., Molchanov, P., Wang, Y.-C.F., Cheng, K.-T., Chen, M.-H.: DoRA: Weight-Decomposed Low-Rank Adaptation (2024). https: //arxiv.org/abs/2402.09353

    2024 arXiv

  39. [47]

    Pratten, G., Garc ´ ıa-Quir´ os, C., Colleoni, M., Ramos-Buades, A., Estell´ es, H., Mateu-Lucena, M., Jaume, R., Haney, M., Keitel, D., Thompson, J.E., Husa, S.: Computationally efficient models for the dominant and subdominant harmonic modes of precessing binary black holes....

    2021 doi

  40. [48]

    Sch¨ afer, M.B., Zelenka, O.c.v., Nitz, A.H., Ohme, F., Br¨ ugmann, B.: Training strategies for deep learning gravitational-wave searches. Phys. Rev. D105, 043002 (2022) https://doi.org/10.1103/PhysRevD.105.043002

    2022 doi

  41. [49]

    IEEE Transactions on Audio and Electroacoustics 15(2), 70–73 (1967) https://doi.org/ 10.1109/TAU.1967.1161901

    Welch, P.: The use of fast fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms. IEEE Transactions on Audio and Electroacoustics 15(2), 70–73 (1967) https://doi.org/ 10.1109/TAU.1967.1161901

    1967

  42. [50]

    Zelenka, O.c.v., Br¨ ugmann, B., Ohme, F.: Convolutional neural networks for signal detection in real ligo data. Phys. Rev. D 110, 024024 (2024) https: //doi.org/10.1103/PhysRevD.110.024024

    2024 doi

  43. [51]

    Koloniari, A.E., Koursoumpa, E.C., Nousi, P., Lampropoulos, P., Passalis, N., Tefas, A., Stergioulas, N.: New gravitational wave discoveries enabled by machine learning. Mach. Learn. Sci. Tech. 6(1), 015054 (2025) https://doi.org/10.1088/ 2632-2153/adb5ed arXiv:2407.07820 [gr-qc]

    2025 arXiv

  44. [52]

    https://doi.org/10.5281/zenodo.4556907

    Nitz, A., Harry, I., Brown, D., Biwer, C.M., Willis, J., Canton, T.D., Capano, C., Pekowsky, L., Dent, T., Williamson, A.R., Davies, G.S., De, S., Cabero, M., Machenschalk, B., Kumar, P., Reyes, S., Macleod, D., dfinstad, Pannarale, F., Massinger, T., Kumar, S., T´ apai, M., S...

    doi

  45. [53]

    Klimenko, S., Vedovato, G., Drago, M., Salemi, F., Tiwari, V., Prodi, G.A., Laz- zaro, C., Ackley, K., Tiwari, S., Da Silva, C.F., Mitselmakher, G.: Method for detection and reconstruction of gravitational wave transients with networks of advanced detectors. Phys. Rev. D 93, 0...

    2016

  46. [54]

    https://arxiv.org/ abs/2501.13846

    Nagarajan, N., Messenger, C.: Identifying and Mitigating Machine Learning Biases for the Gravitational-wave Detection Problem (2025). https://arxiv.org/ abs/2501.13846

    2025

  47. [55]

    Collaboration, T.L.S., Virgo Collaboration, KAGRA Collaboration, Abbott, R., Abe, H., Acernese, F., Ackley, K., Adhicary, S., Adhikari, N., Adhikari, R.X., Adkins, V.K., Adya, V.B., Affeldt, C., Agarwal, D., Agathos, M., Aguiar, O.D., Aiello, L., Ain, A., Ajith, P., Akutsu, T....

    2023

  48. [56]

    Classical and Quantum Gravity 41(8), 085007 (2024) https://doi.org/10.1088/1361-6382/ ad2194

    ´Alvarez-L´ opez, S., Liyanage, A., Ding, J., Ng, R., McIver, J.: Gspynettree: a signal-vs-glitch classifier for gravitational-wave event candidates. Classical and Quantum Gravity 41(8), 085007 (2024) https://doi.org/10.1088/1361-6382/ ad2194

    2024 doi

  49. [57]

    In: Pereira, F., Burges, C.J., Bottou, L., Weinberger, K.Q

    Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep con- volutional neural networks. In: Pereira, F., Burges, C.J., Bottou, L., Weinberger, K.Q. (eds.) Advances in Neural Information Processing Systems, vol. 25. Curran Associates, Inc., ??? (2012)

    2012

  50. [58]

    https://github.com/ML4GW/ml4gw

  51. [59]

    https: //arxiv.org/abs/2407.00143

    Rusak, E., Reizinger, P., Juhos, A., Bringmann, O., Zimmermann, R.S., Brendel, W.: InfoNCE: Identifying the Gap Between Theory and Practice (2025). https: //arxiv.org/abs/2407.00143

    2025 arXiv

  52. [60]

    https://github.com/gwastro/ml-mock-data-challenge-1

  53. [61]

    https://github.com/ondrzel/ml-gw-search/tree/main

  54. [62]

    https://github.com/ondrzel/ml-gw-search/tree/main 31

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.