Pith. sign in

REVIEW 3 major objections 5 minor 29 references

flashcurve: A machine-learning approach for the simple and fast generation of adaptive-binning light curves with Fermi-LAT data

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A convolutional neural network trained on binned Fermi-LAT photon images can estimate detection significance fast enough to make adaptive-binning light curves practical, cutting generation from days to hours.

desk verdict A useful and honest ML tool for fast Fermi-LAT adaptive light curves, but the accuracy claims need quantitative validation and an explicit test on off-grid bin durations. read the letter →

arxiv 2411.12598 v3 pith:7POOWWJF submitted 2024-11-19 astro-ph.IM astro-ph.HE

Theo Glauch , Kristian Tchiorniy This is my paper · ORCID
classification astro-ph.IMastro-ph.HE
keywords Machinelearninggammaraysgalaxies:lightcurvesactiveBLLacertaeobjects:generalmethods:dataanalysisastronomicaldatabases:miscellaneousFermi-LAT
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

Adaptive-binning light curves require many evaluations of how significant a source's gamma-ray emission is in a trial time window; doing these with the full Fermi-LAT likelihood is accurate but slow enough that a single light curve takes hours to days. This paper claims that a convolutional neural network trained on about 1.5 million time bins from the Fermi-LAT Light Curve Repository can estimate the test statistic directly from binned photon images, fast enough to drive a search over time windows. The result would be constant-significance light curves with no upper limits, produced in minutes to a few hours, and the approach is presented as a prototype for machine-learned adaptive binning in other messengers and wavelengths. A sympathetic reader would care because fast adaptive light curves matter for multi-messenger follow-up and for monitoring variable blazars.

What carries the argument

The central object is a residual convolutional neural network (ResNet) that maps a $56 \times 56 \times 6$ image -- photon counts binned in two tangential sky coordinates, with six energy channels spanning $10^2$ to $3\times10^5$ MeV -- to a single positive number, the predicted TS. The skip-connection residual blocks keep the roughly 400,000-parameter network trainable on the 1.5 million time bins. A companion time-window search algorithm scans photon timestamps, predicts TS for each candidate interval, and chooses the last timestamp whose TS falls in a target range (for example [4,9]) as the bin end, avoiding the spuriously short bins that choosing the first in-range timestamp would create.

What would settle it

Assemble test images from time bins whose durations are not 3, 7, or 30 days (for example 10 days and 2 days), compute their true TS with the full Fermi-LAT likelihood analysis, and compare with flashcurve's predictions; a systematic bias that grows with the distance from the three training durations would show the estimator has not generalized to the arbitrary interval lengths the search algorithm produces.

Watch

Extended reading notes

Core claim

The paper's central claim is that the Fermi-LAT detection significance, expressed as the likelihood-ratio test statistic $TS = -2 \ln(L_{\mathrm{max},0}/L_{\mathrm{max},1})$, can be predicted by a convolutional neural network from a four-dimensional histogram of photon counts in time, energy, and two tangential sky coordinates, without solving the full likelihood. Trained on 1.5 million cleaned 3-day, 7-day, and 30-day bins from 1362 variable sources, and validated on four example sources, the network's predictions are accurate enough for a time-bin search algorithm to produce adaptive light curves whose bins have roughly constant TS. The paper argues that this replaces the computationally dominant step of the classic adaptive-binning method, reducing light-curve generation from days to minutes or hours and avoiding upper limits in quiet periods.

Load-bearing premise

The estimator is trained only on time bins of exactly 3, 7, and 30 days, but the search algorithm asks it to judge intervals of arbitrary length without giving the duration as an input; if it has not learned to extrapolate duration scaling, predictions on off-grid interval lengths could be biased and the adaptive light curves would be distorted.

Editorial extensions

If this is right

  • Adaptive-binning Fermi-LAT light curves can be produced in minutes to hours rather than days, making large-scale monitoring and real-time follow-up feasible.
  • Constant-significance bins mean quiet periods get long bins instead of upper limits, so the light curve represents non-detections as real flux measurements.
  • Because prediction is a single forward pass, the same estimator can be reused for any source in the training selection without re-running the likelihood analysis.
  • The approach can be retrained on other data sets, suggesting adaptive binning for other messengers or wavelengths if enough labeled time bins exist.
  • The network's known failure modes, such as source confusion near bright neighbors, are inherited from the image representation and could be addressed by adding neighboring-source information as an extra channel.

Reading between the lines

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

  • The strongest unstated consequence is that the method's accuracy depends on the estimator's ability to extrapolate to interval durations it never saw; the paper does not demonstrate this directly, so an off-grid duration test would be the natural next validation.
  • Because the search algorithm only considers intervals ending at photon timestamps, the resulting time bins are coarser than the full likelihood method's; this is a deliberate trade-off that could be relaxed with a finer or bisection-based search.
  • The approach is framed as a prototype; if it holds, similar convolutional estimators could be trained for other instruments or messengers whose likelihoods are too expensive to evaluate repeatedly, as long as labeled binned data exist.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript presents flashcurve, a convolutional neural network trained to predict the test statistic (TS) of a Fermi-LAT source from binned photon count images, and uses these predictions to drive an adaptive-binning light-curve algorithm. Training data consist of ~1.5 million bins from the Fermi-LAT Light Curve Repository, with fixed durations of 3, 7, and 30 days, represented as 56x56x6 images in right ascension, declination, and six energy bands. The time-bin search (Section 3.6) iteratively extends intervals from photon timestamps until the predicted TS enters a target range, then recomputes fluxes with Fermipy. The authors demonstrate example light curves for four blazars and report runtime improvements of about an order of magnitude or more over the Lott et al. (2012) method.

Significance. If the CNN's TS estimates are reliable at arbitrary interval durations, flashcurve would be a valuable tool for fast adaptive binning, enabling real-time multi-messenger follow-up and large-scale light-curve production. The paper has notable strengths: it uses a large public training dataset, provides open-source code, and includes independent Fermipy cross-checks of flagged outliers (Section 3.5). The central limitation is that generalization to off-grid time durations is asserted but not validated, and the quantitative performance metrics needed to support the constant-significance claim are missing.

major comments (3)
  1. [3.2 / 3.6] The central claim of constant-significance adaptive binning depends on the CNN's ability to predict TS for time intervals of arbitrary duration, but the training data (Section 2.1) contain only bins of 3, 7, and 30 days, and the input representation (Section 3.2) includes no duration, exposure, or expected-background feature. The search algorithm in Section 3.6 evaluates intervals whose endpoints are arbitrary photon timestamps, producing continuous durations that generally differ from the three training values. Because the likelihood TS of a given photon pattern depends on interval duration through the expected background and exposure, the network must interpolate or extrapolate in duration with no explicit input; the paper presents no validation of predictions at off-grid durations. The relative-TS-deviation histograms in Figure 20 are aggregated over all durations and do not report bias or scatter as a function of duration. I request a duration-resolved validation (e.g., predicted vs true TS binned in duration, or a duration-marginalized calibration broken down by duration), or an explicit duration/exposure input feature and retraining.
  2. [4] The paper's headline promise of light curves with constant significance is not quantitatively established. For the four example sources, no statistic is given for the fraction of produced bins whose true TS (from the Fermi-LAT analysis) falls within the target range listed in Table 3. For CTA 102, the text acknowledges that flashcurve frequently under-predicts TS during flaring periods, and Figure 21 suggests that the true TS of flashcurve bins deviates systematically from the target range (50-75) in those periods. To support the constant-significance claim, please report the distribution of true TS and the fraction of bins within (or within a tolerance of) the target range for each example source, and discuss how the search-algorithm thresholds (energy and proximity) contribute to any deviations.
  3. [3.4] The test-set evaluation (Figure 9) is shown only as binned medians and quantiles, and the text claims 'approximately constant relative uncertainty across the entire TS space' without quantitative support. Since the adaptive-binning algorithm selects bins based on predicted TS thresholds (e.g., [4,9] in Section 3.6), the relevant performance is the accuracy and calibration of predicted TS in that low-TS regime. Please provide numeric bias and scatter (e.g., mean/median of predicted-minus-true, RMS, and a calibration plot of predicted TS versus the fraction of true TS exceeding it) for the test set, at minimum in the TS ranges used by the search algorithm.
minor comments (5)
  1. [3.2] The sentence describing energy bin edges as 'half powers of 10 MeV' lists seven edges (10^2 through 10^4.5 and 3x10^5 MeV); the final edge is not a half power of 10, so clarify the bin edges (e.g., 10^2, 10^2.5, ..., 10^5.5).
  2. [4] The comparison to Lott et al. (2012) uses different energy ranges (flashcurve uses 100 MeV-300 GeV, while the Lott method uses an energy cut at E_min) and different target definitions (constant relative flux error vs TS range). Please state explicitly whether the differences in Figures 19-22 could be affected by this mismatch, and consider showing flashcurve with an E_min cut for a more direct comparison.
  3. [Figure 20] Define the 'relative TS deviation' metric in the caption or text (e.g., (TS_pred - TS_true)/TS_true or log ratio).
  4. [3.3.2] The total number of trainable parameters (~400,000) is stated, but the exact number of training epochs and the early-stopping criterion are not given; please add these details for reproducibility.
  5. [3.5] The expected number of outliers per source (~10) is estimated from the flux and flux uncertainty distributions; this estimate deserves a brief derivation or reference.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the network's TS labels are computed by an independent Fermi-LAT likelihood analysis, and the produced light-curve bins are validated with Fermipy, so no prediction reduces to a fitted input by construction.

full rationale

The paper's derivation chain is a supervised regression from raw photon images to a test statistic label taken from the Fermi-LAT Light Curve Repository likelihood analysis (Eq. 3). The labels are produced by an independent maximum-likelihood fit, not by the network, and the trained estimator is evaluated on a held-out test set and on Fermipy-recomputed outlier bins. The time-bin search algorithm in Section 3.6 consumes predicted TS values to choose interval endpoints, but the final light-curve bins are then run through Fermipy to obtain the true TS, so the final significance values are not forced by the network's predictions. The comparison with Lott et al. (2012) sets target TS ranges from that method's output, but this is a comparison/calibration choice, not an input to the network's training or a component of the TS regression itself. The authors' self-citations (Glauch et al. 2022; Giommi et al. 2020; Padovani et al. 2018; Kronmueller and Glauch 2019) are contextual and are not load-bearing for the central claim. The concern that the network was trained only on 3-, 7-, and 30-day bins while the search evaluates arbitrary intervals is a potential generalization/validation gap, but it is not circular reasoning: no fitted parameter of the final light curve is renamed as a prediction, and no equation reduces to its own input by construction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on supervised learning against LCR TS labels and on several transfer assumptions. The model's 400k weights are fit to 1.5 million labels, which is standard training rather than a physical free parameter; the parameters listed here are hand-chosen design choices that affect the result. No new physical entities are introduced.

free parameters (4)
  • Target TS range for search algorithm = [4,9] default; per-source ranges in Table 3 (e.g., 25-50 for TXS 0506+056)
    Hand-chosen threshold controlling bin lengths; affects the resulting light curves and the comparison with Lott et al.
  • Energy bin edges and ROI angular sizes = 6 bins from 10^2 to 3e5 MeV; ROIs 12, 5, 3, 1.5, 1, 0.6 deg
    Hand-chosen input representation following PSF containment; changes what the network sees and therefore its predictions.
  • Network hyperparameters = 15/5/1 residual blocks with 32/64/128 filters; batch size 256
    Chosen such that parameter count is one order of magnitude below sample count; no architecture search is shown.
  • Energy and proximity thresholds in time bin search = not specified numerically
    Used to speed the search by selecting photons with large TS impact; directly affects bin boundary choices and can filter intermediate events.
assumptions (5)
  • domain assumption LCR test statistic values are accurate ground truth for source detection significance.
    The network is trained and tested against these labels; if the LCR likelihood analysis is biased, the estimator inherits the bias. The paper checks some outliers with Fermipy but does not audit all 1.5 million bins.
  • standard math Wilks theorem applies, so TS is approximately chi-square and sqrt(TS) approximates significance.
    Used in Section 2.1 to interpret TS as detection significance. Standard asymptotic result, valid under regularity conditions.
  • ad hoc to paper A CNN can learn TS from binned photon counts in a cone without an explicit exposure map, diffuse background model, or neighbor source list.
    The input is six energy channels of counts; no instrument exposure or background model is provided. The source-confusion failure in Section 3.5 shows this assumption breaks in crowded fields.
  • ad hoc to paper Training on fixed 3/7/30-day bins transfers to arbitrary time intervals produced by the search algorithm.
    Section 3.6 evaluates intervals ending at photon timestamps of any duration, but training data only contain the three fixed durations and duration is not an input feature.
  • domain assumption The training sample of variable extragalactic sources represents all sources for which adaptive light curves will be generated.
    The dataset excludes sources within 10 degrees of the Galactic plane, extended sources, and non-variable sources (Section 2.2). The authors list this as a future improvement, but current claims are limited to the selected population.

how reviews work

0 comments
Cite this review

Pith. "Pith review of flashcurve: A machine-learning approach for the simple and fast generation of adaptive-binning light curves with Fermi-LAT data." pith.science (2026). https://pith.science/paper/7POOWWJF

@misc{pith2026241112598,
  author       = {Pith},
  title        = {Pith review of: flashcurve: A machine-learning approach for the simple and fast generation of adaptive-binning light curves with Fermi-LAT data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7POOWWJF}},
  note         = {Machine review of arXiv:2411.12598}
}
read the original abstract

Gamma rays measured by the Fermi-LAT satellite tell us a lot about the processes taking place in high-energetic astrophysical objects. The fluxes coming from these objects are, however, extremely variable. Hence, gamma-ray light curves optimally use adaptive bin sizes in order to retrieve most information about the source dynamics and to combine gamma-ray observations in a multi-messenger perspective. However, standard adaptive binning approaches are slow, expensive and inaccurate in highly populated regions. Here, we present a novel, powerful, deep-learning-based approach to estimate the necessary time windows for adaptive binning light curves in Fermi-LAT data using raw photon data. The approach is shown to be fast and accurate. It can also be seen as a prototype to train machine-learning models for adaptive binning light curves for other astrophysical messengers.

Figures

Figures reproduced from arXiv: 2411.12598 by the authors.

Figure 1
Figure 1. An example of an LCR ’sanity check’ for time bins with analyses [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Binned distribution of test statistic from the Fermi LCR time bins [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Example neural network input image for 4FGL J0319.8 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (17 more)
Figure 5
Figure 5. Figure 5: Same as the figure before, but using data from the source 4FGL J2253.9 [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: The basic working principle of a residual block used in ResNets, [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 9
Figure 9. Figure 9: Predicted TS from an unseen test set of images vs. the true TS. To [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 10
Figure 10. Figure 10: Learning curve displaying the output of the mean squared error [PITH_FULL_IMAGE:figures/full_fig_p006_10.png]
Figure 8
Figure 8. Figure 8: Learning curve displaying the output of the mean squared error loss [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 11
Figure 11. Figure 11: Learning curve displaying the output of the mean squared error [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 14
Figure 14. Figure 14: Log-binned counts of flux uncertainty data of 4FGL J1234.0-5735, [PITH_FULL_IMAGE:figures/full_fig_p007_14.png]
Figure 12
Figure 12. Figure 12: Set of TS vs. their corresponding squared ratio of flux/ [PITH_FULL_IMAGE:figures/full_fig_p007_12.png]
Figure 15
Figure 15. Figure 15: Image of binned photon event counts for 4FGL J1234.0-5735 in a 3-day period, corresponding to a TS [PITH_FULL_IMAGE:figures/full_fig_p008_15.png]
Figure 16
Figure 16. Figure 16: Same as the figure above for 4FGL J1311.0 [PITH_FULL_IMAGE:figures/full_fig_p008_16.png]
Figure 17
Figure 17. Figure 17: Visualisation of the steps of the time bin search algorithm with a TS [PITH_FULL_IMAGE:figures/full_fig_p009_17.png]
Figure 18
Figure 18. Figure 18: Visualisation of the steps of the time bin search algorithm in the [PITH_FULL_IMAGE:figures/full_fig_p009_18.png]
Figure 19
Figure 19. Figure 19: Light curves of four test sources comparing [PITH_FULL_IMAGE:figures/full_fig_p011_19.png]
Figure 20
Figure 20. Figure 20: Left plots: Histograms of the time bin lengths using flashcurve (red) and the method of Lott et al. (2012) (blue) for the four test sources. Right plots: Histograms of the relative TS deviation of the TS estimated using flashcurve from the TS calculated using the full…
Figure 21
Figure 21. Figure 21: Left plots: TS calculated using the Fermi-LAT analysis for the time bins found using our flashcurve(red) and the method of Lott et al. (2012) (blue) for the four test sources, with respective time bin lengths. In the case of CTA 102, a few extremely high TS are cut ou…
Figure 22
Figure 22. Figure 22: Best-fit spectral indices for the time bins found using [PITH_FULL_IMAGE:figures/full_fig_p014_22.png]
Figure 23
Figure 23. Figure 23: Continuation of Figure 19. Additional zoomed-in light curves. [PITH_FULL_IMAGE:figures/full_fig_p015_23.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 10 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    , et al., year 2015

    author Ajello, M. , et al., year 2015 . title The Origin of the Extragalactic Gamma-Ray Background and Implications for Dark-Matter Annihilation . journal Astrophys. J. Lett. volume 800 , pages L27 . :10.1088/2041-8205/800/2/L27, http://arxiv.org/abs/1501.05301 arXiv:1501.05301

    arXiv 2015

  3. [3]

    , author Bruel, P

    author Ballet, J. , author Bruel, P. , author Burnett, T.H. , author Lott, B. , author collaboration, T.F.L. , year 2024 . title Fermi large area telescope fourth source catalog data release 4 (4fgl-dr4) . https://arxiv.org/abs/2307.12546, http://arxiv.org/abs/2307.12546 arXiv:2307.12546

    arXiv 2024

  4. [4]

    , author Burnett, T.H

    author Ballet, J. , author Burnett, T.H. , author Digel, S.W. , author Lott, B. , year 2020 . title Fermi large area telescope fourth source catalog data release 2 . https://arxiv.org/abs/2005.11208, http://arxiv.org/abs/2005.11208 arXiv:2005.11208

    arXiv 2020

  5. [5]

    , author Faires, D.J

    author Burden, R.L. , author Faires, D.J. , year 1985 . title Numerical analysis . edition 3 ed., publisher PWS Publishing Company , address Boston

    work page 1985

  6. [6]

    , author Gómez-Vargas, G.A

    author Caron, S. , author Gómez-Vargas, G.A. , author Hendriks, L. , author de Austri, R.R. , year 2018 . title Analyzing gamma rays of the galactic center with deep learning . journal Journal of Cosmology and Astroparticle Physics volume 2018 , pages 058–058 . http://dx.doi.org/10.1088/1475-7516/2018/05/058, :10.1088/1475-7516/2018/05/058

    work page doi:10.1088/1475-7516/2018/05/058 2018

  7. [7]

    , author Zeng, H

    author Chai, J. , author Zeng, H. , author Li, A. , author Ngai, E. , year 2021 . title Deep learning in computer vision: A critical review of emerging techniques and application scenarios . journal Machine Learning with Applications volume 6 , pages 100134 . :10.1016/j.mlwa.2021.100134

    arXiv 2021

  8. [8]

    title Fermi/large area telescope bright gamma-ray source list

    author Fermi-LAT , year 2009 . title Fermi/large area telescope bright gamma-ray source list . journal The Astrophysical Journal Supplement Series volume 183 , pages 46–66 . http://dx.doi.org/10.1088/0067-0049/183/1/46, :10.1088/0067-0049/183/1/46

    doi:10.1088/0067-0049/183/1/46 2009

Show all 29 references
  1. [9]

    title Pass 8: Toward the full realization of the fermi-lat scientific potential

    author Fermi-LAT , year 2013 . title Pass 8: Toward the full realization of the fermi-lat scientific potential . https://arxiv.org/abs/1303.3514, http://arxiv.org/abs/1303.3514 arXiv:1303.3514

    2013 arXiv

  2. [10]

    title Fermi large area telescope fourth source catalog

    author Fermi-LAT , year 2020 . title Fermi large area telescope fourth source catalog . journal The Astrophysical Journal Supplement Series volume 247 , pages 33 . https://dx.doi.org/10.3847/1538-4365/ab6bcb, :10.3847/1538-4365/ab6bcb

    2020 doi

  3. [11]

    title The fermi-lat lightcurve repository*

    author Fermi-LAT , year 2023 . title The fermi-lat lightcurve repository* . journal The Astrophysical Journal Supplement Series volume 265 , pages 31 . http://dx.doi.org/10.3847/1538-4365/acbb6a, :10.3847/1538-4365/acbb6a

    2023 doi

  4. [12]

    , author Jacobs, C

    author Fluke, C.J. , author Jacobs, C. , year 2020 . title Surveying the reach and maturity of machine learning and artificial intelligence in astronomy . journal Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery volume 10 , pages e1349

    2020

  5. [13]

    , author Engel, R

    author Gaisser, T.K. , author Engel, R. , author Resconi, E. , year 2016 . title Cosmic Rays and Particle Physics . edition 2 ed., publisher Cambridge University Press

    2016

  6. [14]

    , author Glauch, T

    author Giommi, P. , author Glauch, T. , author Padovani, P. , author Resconi, E. , author Turcati, A. , author Chang, Y.L. , year 2020 . title Dissecting the regions around IceCube high-energy neutrinos: growing evidence for the blazar connection . journal Monthly Notices of t...

    2020 doi

  7. [15]

    , author Kerscher, T

    author Glauch, T. , author Kerscher, T. , author Giommi, P. , year 2022 . title Blast - a machine-learning estimator for the synchrotron peak of blazars . journal Astronomy and Computing volume 41 , pages 100646 . https://www.sciencedirect.com/science/article/pii/S221313372200...

    2022

  8. [16]

    , author Zhang, X

    author He, K. , author Zhang, X. , author Ren, S. , author Sun, J. , year 2015 . title Deep residual learning for image recognition . https://arxiv.org/abs/1512.03385, http://arxiv.org/abs/1512.03385 arXiv:1512.03385

    2015 arXiv

  9. [17]

    , author Roos, M

    author James, F. , author Roos, M. , year 1975 . title Minuit: A System for Function Minimization and Analysis of the Parameter Errors and Correlations . journal Comput. Phys. Commun. volume 10 , pages 343--367 . :10.1016/0010-4655(75)90039-9

    1975 doi

  10. [18]

    , year 2018

    author Kaur, Navpreet , author Baliyan, Kiran S. , year 2018 . title Cta 102 in exceptionally high state during 2016–2017 . journal A and A volume 617 , pages A59 . https://doi.org/10.1051/0004-6361/201731953, :10.1051/0004-6361/201731953

    2018 doi

  11. [19]

    , author Ba, J

    author Kingma, D.P. , author Ba, J. , year 2017 . title Adam: A method for stochastic optimization . https://arxiv.org/abs/1412.6980, http://arxiv.org/abs/1412.6980 arXiv:1412.6980

    2017 arXiv

  12. [20]

    , author Glauch , T

    author Kronmueller , M. , author Glauch , T. , year 2019 . title Application of Deep Neural Networks to Event Type Classification in IceCube . journal arXiv e-prints , pages arXiv:1908.08763 :10.48550/arXiv.1908.08763, http://arxiv.org/abs/1908.08763 arXiv:1908.08763

    2019 doi arXiv

  13. [21]

    , author Bengio, Y

    author LeCun, Y. , author Bengio, Y. , author Hinton, G. , year 2015 . title Deep learning . journal Nature volume 521 , pages 436--444

    2015

  14. [22]

    , author Escande, L

    author Lott, B. , author Escande, L. , author Larsson, S. , author Ballet, J. , year 2012 . title An adaptive-binning method for generating constant-uncertainty/constant-significance light curves withfermi-lat data . journal Astronomy and Astrophysics volume 544 , pages A6 . h...

    2012 doi

  15. [23]

    , author Giommi, P

    author Padovani, P. , author Giommi, P. , author Resconi, E. , author Glauch, T. , author Arsioli, B. , author Sahakyan, N. , author Huber, M. , year 2018 . title Dissecting the region around IceCube-170922A: the blazar TXS 0506+056 as the first cosmic neutrino source . journa...

    2018 doi

  16. [24]

    , author Wang, J

    author Perez, L. , author Wang, J. , year 2017 . title The effectiveness of data augmentation in image classification using deep learning . https://arxiv.org/abs/1712.04621, http://arxiv.org/abs/1712.04621 arXiv:1712.04621

    2017 arXiv

  17. [25]

    , author Foschini, L

    author Sbarrato, T. , author Foschini, L. , author Ghisellini, G. , author Tavecchio, F. , year 2011 . title Study of the variability of blazars gamma-ray emission . journal Advances in Space Research volume 48 , pages 998--1003 . https://www.sciencedirect.com/science/article/...

    2011 doi

  18. [26]

    , author Agarwal, S

    author Sen, S. , author Agarwal, S. , author Chakraborty, P. , author Singh, K.P. , year 2022 . title Astronomical big data processing using machine learning: A comprehensive review . journal Experimental Astronomy volume 53 , pages 1--43

    2022

  19. [27]

    , year 1938

    author Wilks, S.S. , year 1938 . title The Large-Sample Distribution of the Likelihood Ratio for Testing Composite Hypotheses . journal Annals Math. Statist. volume 9 , pages 60--62 . :10.1214/aoms/1177732360

    1938

  20. [28]

    , author Han, N

    author Zhang, X. , author Han, N. , author Zhang, J. , year 2024 . title Comparative analysis of vgg, resnet, and googlenet architectures evaluating performance, computational efficiency, and convergence rates . journal Applied and Computational Engineering volume 44 , pages 1...

    2024 doi

  21. [29]

    , author Wang, L

    author Zhao, X. , author Wang, L. , author Zhang, Y. , author Han, X. , author Deveci, M. , author Parmar, M. , year 2024 . title A review of convolutional neural networks in computer vision . journal Artificial Intelligence Review volume 57 , pages 99 . https://doi.org/10.100...

    2024 doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.