Pith. sign in

REVIEW 5 major objections 5 minor 64 references

Quantum Machine Learning for Identifying Transient Events in X-ray Light Curves

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read A quantum long short-term memory model trained on simulated AGN-like X-ray light curves finds 113 transient-event candidates in the XMM-Newton 4XMM-DR14 catalog, about 1.3 times the 85 found by a classical LSTM with the same…

desk verdict A transparent first application of QLSTM to X-ray transient detection, but the claimed quantum advantage over a classical LSTM is not statistically supported. read the letter →

arxiv 2507.05589 v1 pith:KYBSYSRM submitted 2025-07-08 astro-ph.HE quant-ph

classification astro-ph.HEquant-ph
keywords quantummachinelearningLSTManomalydetectionX-raylightcurvestransienteventsquasi-periodiceruptions4XMM-DR14XMM-Newton
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

The paper asks whether a quantum version of a long short-term memory (LSTM) network can identify transient events in X-ray light curves more effectively than a classical LSTM, and then applies the best quantum model to real data. The authors train on simulated active-galactic-nucleus-like light curves, grade the models on simulated quasi-periodic-eruption flares, and report that quantum LSTMs using superposition and entanglement have higher expressive power and true-positive rates than the classical LSTM at a similar false-positive rate. On 40,154 XMM-Newton pn light curves from 4XMM-DR14, the best QLSTM flags 113 anomaly candidates, about 1.3 times the 85 found by the classical model, and 12 of these have no counterpart in the searched catalogs. The broader point is that quantum machine learning can make a small but concrete difference in an astronomical search, while the candidate list itself is a first step toward finding new X-ray transients.

What carries the argument

The central object is the quantum variational circuit (QVC), a small parameterized quantum circuit that replaces the fully connected layer inside each LSTM gate. Input data are encoded by a feature map; the best-performing map here, ZZFeatureMap, creates Hadamard superpositions, rotates qubits according to input values, and uses CNOT gates to entangle qubits, producing encodings that are hard to reproduce classically. A trainable ansatz then transforms the state; the C20 ansatz entangles neighboring qubits and rotates each qubit around three axes. Pauli-Z measurements on the qubits give the gate outputs, and the authors compare circuits with and without CNOT entanglements to isolate the effect of entanglement. They also measure expressive power with the effective dimension, which counts how many parameters actively shape the model's output for a given sample size.

What would settle it

Use the five known-QPE light curves that pass the paper's filtering (or a larger independently labeled set of real X-ray transients inside 4XMM-DR14) and count how many exceed the threshold-3.0 residual for the trained QLSTM versus the trained CLSTM; if the QLSTM does not recover more real events than the CLSTM at the same false-positive rate, the claimed quantum advantage is a simulation artifact rather than a detection fact.

Watch

Extended reading notes

Core claim

A quantum LSTM built by replacing the fully connected layers of each LSTM gate with quantum variational circuits can serve as an anomaly detector for X-ray time series. Trained only on simulated AGN-like variability and tested on single QPE-like flares, the QLSTM with the ZZFeatureMap and the C20 ansatz achieves a higher effective dimension, comparable false-positive rates, and consistently higher true-positive rates than a classical LSTM with the same parameter count; the advantage survives simulated device noise. When applied to 40,154 good-quality pn light curves from 4XMM-DR14 with a detection threshold of 3.0, this model identifies 113 light curves with anomalous flares, of which 101 have possible counterparts (mostly stellar activity) and 12 have no counterpart in SIMBAD or the ROSAT, SkyMapper, Pan-STARRS, and WISE surveys; one of the unmatched objects shows QPE-like flares. The authors also show the model recovers two of the five available light curves of the known QPE source RX J1301.9+2747.

Load-bearing premise

The comparison stands or falls on whether the computer-generated 'normal' AGN-like light curves and single QPE-like flares behave statistically like real X-ray sources; if real variability is different, the measured false-positive rate, the threshold of 3.0, and the 113-versus-85 gap could all change.

Editorial extensions

If this is right

  • At threshold 3.0 the QLSTM reaches a true-positive rate near 0.8 for simulated flares of amplitude 1.5 while keeping the false-positive rate around 1e-3, so the same procedure should be usable as an automated transient filter on other X-ray catalogs.
  • On 4XMM-DR14 the QLSTM yields 113 transient-event candidates versus 85 for the CLSTM, meaning the quantum model adds about 28 light curves worth examining at the same false-alarm level.
  • The 12 candidates with no catalog counterpart, including the QPE-like object 4XMM J154910.4-534225, are the most promising targets for follow-up observation.
  • Quantum noise from a current-device-derived noise model lowers prediction accuracy by only about 0.01 in loss, and the anomaly-detection ranking is largely preserved, so near-term noisy hardware does not by itself erase the quantum model's advantage.
  • The recovery of two RX J1301.9+2747 light curves among the known QPE sample suggests that the model's sensitivity is adequate for at least the bright, fast class of quasi-periodic eruptions.

Reading between the lines

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

  • At the paper's simulated false-positive rate of about 1e-3, a blind scan of roughly 40,000 light curves would be expected to produce around 40 false alarms, so the 113-candidate list is probably a mixture of real and spurious events rather than a clean catalog.
  • The quantum-over-classical advantage is established on simulated single-flare anomalies; a direct test that injects flares into real light curves, or that runs both models on a labeled set of real transients, would show whether the 1.3-times yield survives outside simulation.
  • The two recovered RX J1301.9+2747 observations suggest the same pipeline could be run on future XMM-Newton releases or on other soft-X-ray missions as a quasi-periodic-eruption candidate finder, an application the paper does not pursue.
  • Because the QLSTM's extra detections are marginal at the chosen threshold, a slightly different threshold choice could erase or enlarge the quantum advantage; reporting the full threshold curve for the catalog run would make the comparison easier to interpret.
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

5 major / 5 minor

Summary. The paper applies a quantum-variant long short-term memory (QLSTM) architecture, in which the fully connected layers of the LSTM gates are replaced by variational quantum circuits, to anomaly detection in X-ray light curves. Normal training data are simulated as AGN-like light curves and anomalous data as AGN-like curves with one injected QPE-like flare. Six QLSTM variants (differing in feature map and ansatz) and a classical LSTM (CLSTM) are trained on the same data. The authors report that QLSTM beats CLSTM in effective dimension, training loss, and true-positive rate, and they select a threshold of 3.0 from a simulated evaluation set (Set B). Applying the best QLSTM to 40,154 4XMM-DR14 pn light curves yields 113 anomaly candidates versus 85 for CLSTM; cross-matching leaves 12 candidates without counterparts. The central claim is that QLSTM 'slightly outperform[s]' CLSTM in expressive power, accuracy, and true-positive rate, and that the higher real-data detection count reflects this advantage.

Significance. If the QLSTM performance advantage were statistically established, this would be a novel and useful demonstration of quantum machine learning in time-domain X-ray astronomy, with a concrete product: a vetted catalog of 113 transient-event candidates and a comparison with a classical baseline. The paper is also commendable for discussing practical limitations of NISQ-based QLSTM (Section 8), documenting the quantum-circuit code (Appendix C), and providing a transparent description of the simulation and thresholding procedure. However, as presented, the central claim is not yet supported: the key comparison in Figure 9 has no uncertainty quantification, the abstract's 'accuracy' claim is contradicted by the validation-loss result, and the same simulated dataset is used for threshold selection and performance assessment. These issues are load-bearing for the paper's main conclusion and for the interpretation of the 113-versus-85 real-data counts.

major comments (5)
  1. [5.3 and Figure 9] The true-positive-rate advantage of QLSTM over CLSTM, which is the main quantitative support for the central claim, is shown only as point estimates with no confidence intervals or significance tests. Set B contains 1000 anomaly light curves per amplitude, so for rates near 0.8 the binomial standard error is about 1.3%; the QLSTM-CLSTM gap in Figure 9 appears to be of this order. Please add error bars (e.g., bootstrap or Wilson intervals), report the exact TPR/FPR values at the adopted threshold of 3.0 for each model, and include a significance test for the QLSTM-versus-CLSTM difference. Without this, the observed gap cannot be distinguished from sampling noise.
  2. [Abstract and Section 7] The abstract claims that QLSTM 'slightly outperform[s]' CLSTM in 'accuracy', but Section 7 explicitly states that for unseen validation data no significant differences are found, and the right panel of Figure 6 shows comparable validation losses. Training loss is not a valid proxy for predictive accuracy. The accuracy claim should be removed or reworded, or the authors should present a validation-based accuracy comparison that supports it. This matters because the abstract's triple claim (expressive power, accuracy, true-positive rate) is the paper's headline.
  3. [5.3 and 6] The threshold of 3.0 is selected using the same Set B that is then used to evaluate the false-positive and true-positive rates and to justify the QLSTM-versus-CLSTM comparison; the best model is also selected using Set A validation loss and Set B ROC-like curves. This creates a selection-bias loop: the reported TPR/FPR may be optimistically biased for the selected model and threshold. I recommend holding out an independent simulated test set (or a labeled real transient sample) for threshold selection, or at minimum reporting a cross-validated or nested evaluation.
  4. [Section 6] The 113-versus-85 counts from 4XMM-DR14 are interpreted as evidence for a QLSTM advantage, but real light curves are unlabeled, and the simulated anomaly training data (single QPE-like flares at fixed amplitudes, Section 3.1) may not represent the diversity of real X-ray transients. The only validation against real known QPEs uses five light curves surviving the filtering, of which two are detected; this is too small and too mixed to establish that the QLSTM finds more genuine transients rather than more false positives. The authors should either add a larger labeled real-transient validation set or explicitly state that the 113/85 comparison is an unvalidated model comparison, not a detection of more real transients.
  5. [3.2] The treatment of real light curves depends on the simulated training distribution (PSD slopes, excess variance range, and QPE flare model), and the final-5 ks exclusion is applied uniformly to all real light curves. If real X-ray variability differs from the simulated AGN-like model, or if genuine flares occur in the excluded final 5 ks, the reported detection counts and the QLSTM-CLSTM gap could change. Please quantify the sensitivity of the 113/85 result to the simulation assumptions, particularly the PSD slope prior and flare amplitude distribution, and justify the uniform 5-ks exclusion with an example or reference.
minor comments (5)
  1. [Figure 4 caption] The caption lists six QVCs as '(C20, C20), (AngleX, C20), (ZZFeatureMap, C20), (ZZFeatureMap, C20), (ZZFeatureMap, C20NoCNOT), and (AngleX, C20)', which duplicates ZZFeatureMap/C20 and omits (ZZFeatureMap, RealAmplitudes); the caption should match Table 1.
  2. [Section 7] The summary text contains the typo 'CLTM' instead of 'CLSTM' in the sentence 'This detection number is about 1.3 times higher than that by the CLTM model.'
  3. [Section 6] There is a typo in the sentence about an anomaly 'mioverlooked by the CLSTM model' that should read 'overlooked'.
  4. [3.2 / Figure 10] The text explains binning of real light curves but not how they are standardized before being fed to the model; since the model was trained on standardized simulated light curves, the standardization procedure for real data should be stated explicitly.
  5. [5.3] The false-positive rate at threshold 3.0 is quoted as approximately 1e-3, but with 1000 normal light curves in Set B the statistical precision of such a low rate is limited; reporting a confidence interval for the FPR would clarify how well this rate is actually constrained.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: model, threshold, and catalog comparison are evaluated out-of-sample on independently drawn simulated data, with no fitted parameter renamed as a prediction.

full rationale

The derivation chain is not circular. The QLSTM and CLSTM models are trained on Set A simulated AGN-like light curves and evaluated on Set B (1000 normal and 1000 anomaly curves) and Set C, which are distinct draws from the same generative process rather than the training subset; no parameter is fitted to the real 4XMM-DR14 data before the models are applied. Defining anomalies as AGN-like variability plus one QPE-like flare is the target concept that supplies labels for a standard anomaly-detection benchmark, not a self-definitional reduction in which the output is equal to the input by construction. The threshold of 3.0 is selected from the Set B false-positive/true-positive curves and then applied identically to both QLSTM and CLSTM on real data, so the 113-vs-85 comparison is a relative benchmark under the same operating point; its absolute value inherits the simulation assumptions, but that is an extrapolation caveat, not circularity. Citations to Yen-Chi Chen et al. (2020) for the QLSTM architecture, Abbas et al. (2020) for effective dimension, and Timmer & König (1995) for light-curve simulation are external to the present author list and are not used as self-citations or uniqueness theorems. The only notable weakness is that the abstract's 'accuracy' component is qualified by the paper's own admission that validation losses show no significant differences, but that is an internal consistency and statistical-significance concern, not a circularity of the derivation.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on simulated-data fidelity, a threshold chosen from the same simulated ROC curves, and standard quantum-computing and ML assumptions. No new physical entities are introduced.

free parameters (3)
  • Detection threshold = 3.0 (standardized residual)
    Set in Section 5.3 from ROC curves on simulated data to balance false- and true-positive rates. Directly determines the 113 detections in Section 6.
  • QPE flare amplitudes = 1.1, 1.5, 2.0, 3.0
    Chosen in Section 3.1 to bracket observed QPE amplitudes; the performance assessment uses amplitudes greater than or equal to 1.5, constraining the claimed QLSTM advantage.
  • Model hyperparameters = hidden size=2, qubits=4, ansatz reps=2 (6 for RealAmplitudes), 50 epochs, learning rate 0.05
    Selected in Section 4.1 to match CLSTM parameter count and computational limits; arbitrary choices that can affect the QLSTM versus CLSTM comparison.
assumptions (5)
  • domain assumption AGN X-ray variability is a power-law PSD process with slopes drawn from Gaussian(mean=-2, std=0.3).
    Section 3.1: normal training data are simulated this way; if real 4XMM-DR14 variability differs, the learned normal model and threshold may not transfer.
  • domain assumption Excess variance is drawn uniformly in log space over [-5,-1].
    Section 3.1: sets variability strength of simulated normal data, directly affecting flare visibility in Figures 7 and 8.
  • domain assumption QPE-like flares follow exponentially modified Gaussian profiles, one per light curve, with parameters from Webbe and Young (2023).
    Section 3.1: defines the anomaly class used for training assessment and threshold selection; real transients may differ in shape, duration, or recurrence.
  • domain assumption A prediction residual exceeding a fixed threshold is a valid indicator of an astrophysical transient, and the final 5 ks of every light curve can be excluded as non-celestial.
    Sections 3.2 and 5.3: underlies the detection method; the final-5 ks cut may remove genuine late-time transients.
  • domain assumption The effective dimension computed with a binary-classification simplification is a valid proxy for expressive power of the LSTM models.
    Section 4.2 and Appendix D: the metric is borrowed from Abbas et al. (2020) and simplified for LSTM; its connection to anomaly-detection performance is indirect.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Machine Learning for Identifying Transient Events in X-ray Light Curves." pith.science (2026). https://pith.science/paper/KYBSYSRM

@misc{pith2026250705589,
  author       = {Pith},
  title        = {Pith review of: Quantum Machine Learning for Identifying Transient Events in X-ray Light Curves},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KYBSYSRM}},
  note         = {Machine review of arXiv:2507.05589}
}
abstract

We investigate whether a novel method of quantum machine learning (QML) can identify anomalous events in X-ray light curves as transient events and apply it to detect such events from the XMM-Newton 4XMM-DR14 catalog. The architecture we adopt is a quantum version of the long-short term memory (LSTM) where some fully connected layers are replaced with quantum circuits. The LSTM, making predictions based on preceding data, allows identification of anomalies by comparing predicted and actual time-series data. The necessary training data are generated by simulating active galactic nucleus-like light curves as the species would be a significant population in the XMM-Newton catalog. Additional anomaly data used to assess trained quantum LSTM (QLSTM) models are produced by adding flares like quasi-periodic eruptions to the training data. Comparing various aspects of the performances of the quantum and classical LSTM models, we find that QLSTM models incorporating quantum superposition and entanglement slightly outperform the classical LSTM (CLSTM) model in expressive power, accuracy, and true-positive rate. The highest-performance QLSTM model is then used to identify transient events in 4XMM-DR14. Out of 40154 light curves in the 0.2--12 keV band, we detect 113 light curves with anomalies, or transient event candidates. This number is $\approx$ 1.3 times that of anomalies detectable with the CLSTM model. By utilizing SIMBAD and four wide-field survey catalogs made by ROSAT, SkyMapper, Pan-STARRS, and WISE, no possible counterparts are found for 12 detected anomalies.

Figures

Figures reproduced from arXiv: 2507.05589 by the authors.

Figure 1
Figure 1. Example of the state of a qubit, which is the fun￾damental unit in quantum computing, on a complex sphere or the Bloch sphere. The magenta arrow indicates the state |Φ⟩ = 1/ √ 2|0⟩+1/ √ 2|1⟩. The initial state of |0⟩ corresponds to the arrow pointing straight upward. quantum ML (QML). As detailed in Appendix A, quan￾tum computing uses a fundamental unit called “qubit”. The qubit is a two-level system, and various qu… view at source ↗
Figure 2
Figure 2. Left: Schematic picture of a single CLSTM layer at a time step of t. The layer takes four inputs of ct−1, ht−1, t, and xt, which are the cell state (long-term memory), hidden state (shot-term memory), time, and a vector of input values at the time, respectively. The FCL denotes a fully connected layer. Right: the schematic picture of a single QLSTM layer at a time step of t. The FCLs of the gates in the CLSTM layer … view at source ↗
Figure 3
Figure 3. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (15 more)
Figure 4
Figure 4. Figure 4: From top to bottom, six types of QVCs are presented in the order of [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Effective dimension divided by the actual dimen￾sion as a function of sample size. The data points and error bars indicate the average values and standard deviations of multiple calculations, respectively. For each QLSTM model, we show the result of 10 calculations, wh…
Figure 6
Figure 6. Figure 6: Loss histories for the training and validation datasets, shown in the top and bottom figures, respectively. We use the Set A data, providing 100 normal light curves (see Section 3.1). For each line, the used feature map and ansatz are presented in the legend. time-seri…
Figure 7
Figure 7. Figure 7: QML predictions for nine representative, simulated AGN-like light curves in Set C (Section 3.1). Results are based on the pivotal QML model constructed with (ZZFeatureMap, C20). The upper sub-panel of each panel shows the normal light curve (gray solid line), the noise…
Figure 8
Figure 8. Figure 8: Same as [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: False-positive and true-positive rates as a func￾tion of the threshold for distinguishing between normal and anomalous events. The top and middle panels display the results for the amplitudes of 1.1 and 1.5, respectively. The bottom panel replicates the middle one but …
Figure 10
Figure 10. Figure 10: Standardized XMM-Newton light curves in the 0.2–12 keV band (blue) and QLSTM model predictions (orange) using ZZFeatureMap and C20. These light curves lack possible counterparts in the referenced databases. Lower panels show the absolute differences between the real a…
Figure 11
Figure 11. Figure 11: Standardized XMM-Newton light curve in the 0.2–12 keV band (blue), and light curves predicted by the best QLSTM model and the CLSTM one. The lower panel shows the absolute differences between the real and predicted values, with orange dashed and gray dot-dashed lines …
Figure 12
Figure 12. Figure 12: The leftmost sphere indicates the |0⟩ state of a qubit in the Bloch sphere. The arrows in the other spheres from left to right show the results after applying the RY(π/2), RZ(π/2), and RX(−π/2) gates to the qubit of |0⟩ sequentially. The final state of the qubit shown…
Figure 13
Figure 13. Figure 13: Top two rows: the prediction results for the light curve simulated with σ 2 NXS = 10−3 and the PSD index of −2.0 using the CLSTM and five QLSTM models excluding the fiducial QLSTM model. The fluxes are represented in arbitrary units. Bottom two rows: the results for t…
Figure 14
Figure 14. Figure 14: Standardized XMM-Newton light curves in the 0.2–12 keV band (blue) and predicted ones (orange) based on the QLSTM model constructed with ZZFeatureMap and C20. The lower panels show the absolute values of the differences between the real and predicted values. In each f…
Figure 15
Figure 15. Figure 15: Standardized XMM-Newton light curves in the 0.2–12 keV band (blue) and predicted ones (orange) based on the QLSTM model constructed with ZZFeatureMap and C20. The lower panels show the absolute values of the differences between the real and predicted values. In each f…
Figure 16
Figure 16. Figure 16: Standardized XMM-Newton light curves in the 0.2–12 keV band (blue) and predicted ones (orange) based on the QLSTM model constructed with ZZFeatureMap and C20. The lower panels show the absolute values of the differences between the real and predicted values. In each f…
Figure 17
Figure 17. Figure 17: Standardized XMM-Newton light curves in the 0.2–12 keV band (blue) and predicted ones (orange) based on the QLSTM model constructed with ZZFeatureMap and C20. The lower panels show the absolute values of the differences between the real and predicted values. In each f…
Figure 18
Figure 18. Figure 18: Standardized XMM-Newton light curves in the 0.2–12 keV band (blue) and predicted ones (orange) based on the QLSTM model constructed with ZZFeatureMap and C20. The lower panels show the absolute values of the differences between the real and predicted values. In each f…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 10 canonical work pages

  1. [1]

    K., Ag¨ ueros, M

    Abazajian, K., Adelman-McCarthy, J. K., Ag¨ ueros, M. A., et al. 2003, AJ, 126, 2081, doi: 10.1086/378165

  2. [2]

    2020, arXiv e-prints, arXiv:2011.00027, doi: 10.48550/arXiv.2011.00027

    Abbas, A., Sutter, D., Zoufal, C., et al. 2020, arXiv e-prints, arXiv:2011.00027, doi: 10.48550/arXiv.2011.00027

  3. [3]

    2020, ApJ, 896, 39, doi: 10.3847/1538-4357/ab91ba

    Alp, D., & Larsson, J. 2020, ApJ, 896, 39, doi: 10.3847/1538-4357/ab91ba

  4. [4]

    K., & Thorat, K

    Aniyan, A. K., & Thorat, K. 2017, ApJS, 230, 20, doi: 10.3847/1538-4365/aa7333

  5. [5]

    2021, Nature, 592, 704, doi: 10.1038/s41586-021-03394-6 Astropy Collaboration, Robitaille, T

    Arcodia, R., Merloni, A., Nandra, K., et al. 2021, Nature, 592, 704, doi: 10.1038/s41586-021-03394-6 Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33, doi: 10.1051/0004-6361/201322068 Astropy Collaboration, Price-Whelan, A. M., Sip˝ ocz, B. M., et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f

  6. [6]

    2022, MNRAS, 509, 3504, doi: 10.1093/mnras/stab3259

    Bao, T., & Li, Z. 2022, MNRAS, 509, 3504, doi: 10.1093/mnras/stab3259

  7. [7]

    H., White, R

    Becker, R. H., White, R. L., & Helfand, D. J. 1995, ApJ, 450, 559, doi: 10.1086/176166

  8. [8]

    2020, arXiv e-prints, arXiv:2001.10872, doi: 10.48550/arXiv.2001.10872

    Berezniuk, O., Figalli, A., Ghigliazza, R., & Musaelian, K. 2020, arXiv e-prints, arXiv:2001.10872, doi: 10.48550/arXiv.2001.10872

Show all 64 references
  1. [9]

    2014, Advances in Space Research, 53, 900, doi: 10.1016/j.asr.2013.07.045

    Bianchi, L., Conti, A., & Shiao, B. 2014, Advances in Space Research, 53, 900, doi: 10.1016/j.asr.2013.07.045

  2. [10]

    J., Geim, A

    Bluvstein, D., Evered, S. J., Geim, A. A., et al. 2024, Nature, 626, 58, doi: 10.1038/s41586-023-06927-3

  3. [11]

    J., Tr¨ umper, J., et al

    Boller, T., Freyberg, M. J., Tr¨ umper, J., et al. 2016, A&A, 588, A103, doi: 10.1051/0004-6361/201525648

  4. [12]

    2024, arXiv e-prints, arXiv:2408.11047, doi: 10.48550/arXiv.2408.11047

    Corli, S., Moro, L., Dragoni, D., Dispenza, M., & Prati, E. 2024, arXiv e-prints, arXiv:2408.11047, doi: 10.48550/arXiv.2408.11047

  5. [13]

    Davenport, J. R. A. 2016, ApJ, 829, 23, doi: 10.3847/0004-637X/829/1/23

  6. [14]

    R., Chornock, R., Soderberg, A

    Drout, M. R., Chornock, R., Soderberg, A. M., et al. 2014, ApJ, 794, 23, doi: 10.1088/0004-637X/794/1/23

  7. [15]

    A., Mahabal, A., Masci, F

    Duev, D. A., Mahabal, A., Masci, F. J., et al. 2019, MNRAS, 489, 3582, doi: 10.1093/mnras/stz2357

  8. [16]

    P., et al

    Fender, R., Stewart, A., Macquart, J. P., et al. 2015, in Advancing Astrophysics with the Square Kilometre Array (AASKA14), 51, doi: 10.22323/1.215.0051

  9. [17]

    A., Magnier, E

    Flewelling, H. A., Magnier, E. A., Chambers, K. C., et al. 2020, ApJS, 251, 7, doi: 10.3847/1538-4365/abb82d Gonz´ alez-Mart ´ ın, O., & Vaughan, S. 2012, A&A, 544, A80, doi: 10.1051/0004-6361/201219008

  10. [18]

    2023, arXiv e-prints, arXiv:2306.14990, doi: 10.48550/arXiv.2306.14990

    Haiman, Z., Xin, C., Bogdanovi´ c, T., et al. 2023, arXiv e-prints, arXiv:2306.14990, doi: 10.48550/arXiv.2306.14990

  11. [19]

    2023, ApJ, 942, 9, doi: 10.3847/1538-4357/aca283

    Hammerstein, E., van Velzen, S., Gezari, S., et al. 2023, ApJ, 942, 9, doi: 10.3847/1538-4357/aca283

  12. [20]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2 Havl ´ ıˇ cek, V., C´ orcoles, A. D., Temme, K., et al. 2019, Nature, 567, 209, doi: 10.1038/s41586-019-0980-2

  13. [21]

    Ho, A. Y. Q., Perley, D. A., Gal-Yam, A., et al. 2023, ApJ, 949, 120, doi: 10.3847/1538-4357/acc533

  14. [22]

    1997, Neural computation, 9, 1735

    Hochreiter, S., & Schmidhuber, J. 1997, Neural computation, 9, 1735

  15. [23]

    2021, Nature Communications, 12, 2631, doi: 10.1038/s41467-021-22539-9

    Huang, H.-Y., Broughton, M., Mohseni, M., et al. 2021, Nature Communications, 12, 2631, doi: 10.1038/s41467-021-22539-9

  16. [24]

    Hunter, J. D. 2007, Computing in Science and Engineering, 9, 90, doi: 10.1109/MCSE.2007.55 Hyper-Kamiokande Proto-Collaboration, :, Abe, K., et al. 2018, arXiv e-prints, arXiv:1805.04163, doi: 10.48550/arXiv.1805.04163 Ivezi´ c,ˇZ., Kahn, S. M., Tyson, J. A., et al. 2019, ApJ,...

  17. [25]

    2001, A&A, 365, L1, doi: 10.1051/0004-6361:20000036

    Jansen, F., Lumb, D., Altieri, B., et al. 2001, A&A, 365, L1, doi: 10.1051/0004-6361:20000036

  18. [26]

    2016, PASJ, 68, 58, doi: 10.1093/pasj/psw056 —

    Kawamuro, T., Ueda, Y., Shidatsu, M., et al. 2016, PASJ, 68, 58, doi: 10.1093/pasj/psw056 —. 2018, ApJS, 238, 32, doi: 10.3847/1538-4365/aad1ef

  19. [27]

    2023, Nature, 618, 500, doi: 10.1038/s41586-023-06096-3 Kovaˇ cevi´ c, M., Pasquato, M., Marelli, M., et al

    Kim, Y., Eddins, A., Anand, S., et al. 2023, Nature, 618, 500, doi: 10.1038/s41586-023-06096-3 Kovaˇ cevi´ c, M., Pasquato, M., Marelli, M., et al. 2022, A&A, 659, A66, doi: 10.1051/0004-6361/202142444

  20. [28]

    R., Harrison, F

    Kulkarni, S. R., Harrison, F. A., Grefenstette, B. W., et al. 2021, arXiv e-prints, arXiv:2111.15608, doi: 10.48550/arXiv.2111.15608

  21. [29]

    Li, Y., & Benjamin, S. C. 2016, arXiv e-prints, arXiv:1611.09301, doi: 10.48550/arXiv.1611.09301

  22. [30]

    D., et al

    Lien, A., Sakamoto, T., Barthelmy, S. D., et al. 2016, ApJ, 829, 7, doi: 10.3847/0004-637X/829/1/7

  23. [31]

    A., & Barret, D

    Lin, D., Webb, N. A., & Barret, D. 2012, ApJ, 756, 27, doi: 10.1088/0004-637X/756/1/27

  24. [32]

    2019, ApJ, 870, 126, doi: 10.3847/1538-4357/aaf39b

    Lin, J., Yan, Z., Han, Z., & Yu, W. 2019, ApJ, 870, 126, doi: 10.3847/1538-4357/aaf39b

  25. [33]

    2020, JCAP, 2020, 050, doi: 10.1088/1475-7516/2020/03/050

    Maggiore, M., Van Den Broeck, C., Bartolo, N., et al. 2020, JCAP, 2020, 050, doi: 10.1088/1475-7516/2020/03/050

  26. [34]

    2011, ApJ, 731, 53, doi: 10.1088/0004-637X/731/1/53

    Mainzer, A., Bauer, J., Grav, T., et al. 2011, ApJ, 731, 53, doi: 10.1088/0004-637X/731/1/53

  27. [35]

    2022, A&A, 664, A81, doi: 10.1051/0004-6361/202142952

    Makhlouf, K., Turpin, D., Corre, D., et al. 2022, A&A, 664, A81, doi: 10.1051/0004-6361/202142952

  28. [36]

    Metzger, B. D. 2017, Living Reviews in Relativity, 20, 3, doi: 10.1007/s41114-017-0006-z

  29. [37]

    Moshir, M., Kopman, G., & Conrow, T. A. O. 1992, IRAS Faint Source Survey, Explanatory supplement version 2 Quantum machine learning for anomaly detection 35

  30. [38]

    S., Lochner, M., Webb, S., & Narayan, G

    Muthukrishna, D., Mandel, K. S., Lochner, M., Webb, S., & Narayan, G. 2022, MNRAS, 517, 393, doi: 10.1093/mnras/stac2582

  31. [39]

    2013, arXiv e-prints, arXiv:1306.2307, doi: 10.48550/arXiv.1306.2307

    Nandra, K., Barret, D., Barcons, X., et al. 2013, arXiv e-prints, arXiv:1306.2307, doi: 10.48550/arXiv.1306.2307

  32. [40]

    2014, ApJ, 793, 23, doi: 10.1088/0004-637X/793/1/23

    Nun, I., Pichara, K., Protopapas, P., & Kim, D.-W. 2014, ApJ, 793, 23, doi: 10.1088/0004-637X/793/1/23

  33. [41]

    2024, ApJS, 272, 13, doi: 10.3847/1538-4365/ad344f

    Panda, S., & ´Sniegowska, M. 2024, ApJS, 272, 13, doi: 10.3847/1538-4365/ad344f

  34. [42]

    L., Lieu, M., & Matzeu, G

    Parker, M. L., Lieu, M., & Matzeu, G. A. 2022, MNRAS, 514, 4061, doi: 10.1093/mnras/stac1639

  35. [43]

    2019, arXiv e-prints, arXiv:1912.01703, doi: 10.48550/arXiv.1912.01703 Peral Garc ´ ıa, D., Cruz-Benito, J., & Jos´ e Garc ´ ıa-Pe˜ nalvo, F

    Paszke, A., Gross, S., Massa, F., et al. 2019, arXiv e-prints, arXiv:1912.01703, doi: 10.48550/arXiv.1912.01703 Peral Garc ´ ıa, D., Cruz-Benito, J., & Jos´ e Garc ´ ıa-Pe˜ nalvo, F. 2022, arXiv e-prints, arXiv:2201.04093, doi: 10.48550/arXiv.2201.04093 P´ erez-D ´ ıaz, V. S.,...

  36. [44]

    2021, ApJ, 913, 60, doi: 10.3847/1538-4357/abf24d S´ anchez-S´ aez, P., Lira, H., Mart ´ ı, L., et al

    Poolakkil, S., Preece, R., Fletcher, C., et al. 2021, ApJ, 913, 60, doi: 10.3847/1538-4357/abf24d S´ anchez-S´ aez, P., Lira, H., Mart ´ ı, L., et al. 2021, AJ, 162, 206, doi: 10.3847/1538-3881/ac1426

  37. [45]

    2015, Contemporary Physics, 56, 172, doi: 10.1080/00107514.2014.964942

    Schuld, M., Sinayskiy, I., & Petruccione, F. 2015, Contemporary Physics, 56, 172, doi: 10.1080/00107514.2014.964942

  38. [46]

    2022, PASJ, 74, 612, doi: 10.1093/pasj/psac023

    Shimakawa, R., Tanaka, M., Bottrell, C., et al. 2022, PASJ, 74, 612, doi: 10.1093/pasj/psac023

  39. [47]

    2024, ApJ, 964, 74, doi: 10.3847/1538-4357/ad2704

    Shvartzvald, Y., Waxman, E., Gal-Yam, A., et al. 2024, ApJ, 964, 74, doi: 10.3847/1538-4357/ad2704

  40. [48]

    S., Shimakawa, R., Shimasaku, K., et al

    Tanaka, T. S., Shimakawa, R., Shimasaku, K., et al. 2022, PASJ, 74, 1, doi: 10.1093/pasj/psab105

  41. [49]

    Team, T. P. D. 2023, pandas-dev/pandas: Pandas, v2.1.1,

  42. [50]

    Zenodo, Zenodo, doi: 10.5281/zenodo.3509134

  43. [51]

    Temme, K., Bravyi, S., & Gambetta, J. M. 2017, PhRvL, 119, 180509, doi: 10.1103/PhysRevLett.119.180509

  44. [52]

    1995, A&A, 300, 707

    Timmer, J., & K¨ onig, M. 1995, A&A, 300, 707

  45. [53]

    2023, MNRAS, 526, 1687, doi: 10.1093/mnras/stad2775

    Tortosa, A., Ricci, C., Ar´ evalo, P., et al. 2023, MNRAS, 526, 1687, doi: 10.1093/mnras/stad2775

  46. [54]

    2022, A&A, 657, A138, doi: 10.1051/0004-6361/202141259

    Tranin, H., Godet, O., Webb, N., & Primorac, D. 2022, A&A, 657, A138, doi: 10.1051/0004-6361/202141259

  47. [55]

    D., Lamer, G., et al

    Traulsen, I., Schwope, A. D., Lamer, G., et al. 2020, A&A, 641, A137, doi: 10.1051/0004-6361/202037706

  48. [56]

    F., Barthelmy, S., et al

    Tueller, J., Mushotzky, R. F., Barthelmy, S., et al. 2008, ApJ, 681, 113, doi: 10.1086/588458

  49. [57]

    A., Cranmer, M., Berger, E., et al

    Villar, V. A., Cranmer, M., Berger, E., et al. 2021, ApJS, 255, 24, doi: 10.3847/1538-4365/ac0893

  50. [58]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, doi: 10.1038/s41592-019-0686-2

  51. [59]

    1999, A&A, 349, 389, doi: 10.48550/arXiv.astro-ph/9909315 von Kienlin, A., Meegan, C

    Voges, W., Aschenbach, B., Boller, T., et al. 1999, A&A, 349, 389, doi: 10.48550/arXiv.astro-ph/9909315 von Kienlin, A., Meegan, C. A., Paciesas, W. S., et al. 2020, ApJ, 893, 46, doi: 10.3847/1538-4357/ab7a18

  52. [60]

    A., Coriat, M., Traulsen, I., et al

    Webb, N. A., Coriat, M., Traulsen, I., et al. 2020, A&A, 641, A136, doi: 10.1051/0004-6361/201937353

  53. [61]

    Webbe, R., & Young, A. J. 2023, RAS Techniques and Instruments, 2, 238, doi: 10.1093/rasti/rzad015

  54. [62]

    A., Luvaul, L

    Wolf, C., Onken, C. A., Luvaul, L. C., et al. 2018, PASA, 35, e010, doi: 10.1017/pasa.2018.5

  55. [63]

    L., Eisenhardt, P

    Wright, E. L., Eisenhardt, P. R. M., Mainzer, A. K., et al. 2010, AJ, 140, 1868, doi: 10.1088/0004-6256/140/6/1868

  56. [64]

    2015, ApJ, 805, 87, doi: 10.1088/0004-637X/805/2/87 Yen-Chi Chen, S., Yoo, S., & Fang, Y.-L

    Yan, Z., & Yu, W. 2015, ApJ, 805, 87, doi: 10.1088/0004-637X/805/2/87 Yen-Chi Chen, S., Yoo, S., & Fang, Y.-L. L. 2020, arXiv e-prints, arXiv:2009.01783, doi: 10.48550/arXiv.2009.01783

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.