Pith. sign in

REVIEW 2 major objections 5 minor 76 references

A CNN trained on simulated Lyα forests recovers absorber properties fast enough for upcoming surveys and still preserves the observed column-density distribution.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 06:50 UTC pith:AEWZA3FQ

load-bearing objection Solid incremental engineering paper: TNG-trained sliding-window CNN with MC-dropout that still recovers CDDF and b–N envelope on UVES despite a clear domain gap. the 2 major comments →

arxiv 2607.05494 v1 pith:AEWZA3FQ submitted 2026-07-06 astro-ph.GA

Uncertainty-Aware Deep Learning for the Lyα Forest: CNN-Based Absorber Detection and Characterization

Paryag Sharma , Vikram Khaire , Ting-Yun Cheng , Hum Chand , Prakash Gaikwad This is my paper
classification astro-ph.GA
keywords Lyα forestconvolutional neural networkVoigt profile fittingcolumn density distribution functionintergalactic mediumdomain shiftMonte Carlo dropoutquasar absorption lines
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Traditional Voigt-profile fitting of the Lyα forest is too slow and human-intensive for the millions of quasar spectra expected from next-generation surveys. This paper shows that a sliding-window convolutional neural network, trained on synthetic spectra drawn from a cosmological simulation and labelled by the same Voigt fitter used on real data, can simultaneously detect absorbers and predict their column density, Doppler width and centroid. On the simulated test set the network reaches an F1 score of about 0.8 and mean absolute errors of 0.18 dex and 0.10 dex in log N_HI and log b_HI; it also recovers the column-density distribution function slope and the lower envelope of the b–N relation to within a few tenths of a km s^{-1}. When the same network is applied to real high-resolution UVES spectra the F1 score falls to ~0.5 and the parameter errors roughly double, yet the statistical distributions that matter for cosmology remain consistent with the classical catalogue. Monte-Carlo dropout supplies per-absorber uncertainties, and inference on ten spectra takes less than a minute on a laptop CPU. The result is a practical, uncertainty-aware pipeline that can keep pace with the coming flood of spectroscopic data.

Core claim

A multi-task sliding-window CNN trained on TNG sightlines labelled by VIPER recovers absorber presence, log N_HI, log b_HI and line centroid with F1 ~0.8 and MAE ~0.18/0.10 on simulated spectra; despite a domain-shift-driven drop to F1 ~0.5 on real UVES data, the network still reproduces the observed CDDF slope and b–N lower envelope (RMS 2.96 km s^{-1}) while running orders of magnitude faster than classical fitting.

What carries the argument

The sliding-window multi-task CNN that maps a 691-pixel flux segment to four simultaneous outputs (binary LyID, log N_HI, log b_HI, centroid offset) and uses Monte-Carlo dropout at inference to furnish epistemic uncertainties.

Load-bearing premise

The assumption that VIPER labels derived from simulated TNG sightlines form an adequate ground-truth distribution for real UVES spectra, even though the paper itself measures a clear latent-space domain shift between the two.

What would settle it

If a larger, independent UVES or HIRES sample analysed with the same VIPER pipeline yields a CDDF slope or b–N lower-envelope RMS difference that is statistically inconsistent with the CNN catalogue once domain-adaptation methods are applied, the claim that the network preserves the key forest statistics collapses.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper presents a multi-task sliding-window CNN that identifies Lyα absorbers and predicts log N_HI, log b_HI, and line centroid from quasar spectra. Training spectra are generated from IllustrisTNG with trident and labelled by VIPER Voigt-profile fits; inference uses Monte Carlo dropout for epistemic uncertainties. On held-out TNG spectra the model reaches F1 ~0.8 with MAE ~0.18 (log N_HI) and ~0.10 (log b_HI), and recovers the CDDF slope and b–N lower envelope to high accuracy. On UVES spectra performance falls to F1 ~0.5 and larger MAEs, which the authors attribute to a measured latent-space domain shift (centroid distance 0.67), yet the observed CDDF slope and lower-envelope shape remain consistent with VIPER. The work positions the CNN as a fast, uncertainty-aware surrogate for large spectroscopic surveys.

Significance. If the reported statistical fidelity holds, the method offers a practical route to absorber catalogues for the millions of spectra expected from DESI and similar surveys, where traditional Voigt fitting is prohibitive. Strengths include the use of hydrodynamical rather than purely analytic training spectra, explicit quantification of domain shift, Monte Carlo dropout uncertainties, and transparent reporting of the simulation-to-observation performance drop. The central claim is appropriately limited to recovery of population statistics rather than per-absorber physical fidelity on real data, which keeps the contribution well-scoped and useful.

major comments (2)
  1. Section 2.3.2 and the evaluation protocol in Section 4: blended systems are labelled by overwriting weaker components with the highest-N_HI absorber, and matching uses a single Δv < 10 km s^{-1} criterion. Because real forests are heavily blended, this rule can systematically bias both training targets and the TP/FP accounting that underlies the reported F1 and MAE. A quantitative test (e.g., recovery statistics stratified by number of VIPER components, or an ablation that retains multi-component labels) is needed to show that the CDDF and lower-envelope agreement on UVES data is not an artefact of this simplification.
  2. Section 5 (latent-space analysis) and Figures 6–9: the Euclidean centroid distance of 0.67 between simulated and UVES latent distributions is correctly diagnosed as the main cause of the F1 drop, yet the paper still presents the UVES CDDF slope (1.39 vs 1.40) and lower-envelope RMS (2.96 km s^{-1}) as primary evidence of scientific utility. Without an uncertainty budget that folds the domain-shift residual into the CDDF and envelope errors, or a domain-adaptation experiment that reduces the latent distance, it remains unclear whether the statistical agreement is robust enough for survey-scale science. A short additional analysis quantifying how much of the residual scatter is attributable to the domain gap would strengthen the claim.
minor comments (5)
  1. Section 3.2: the equal-weight total loss is stated after noting that the individual terms still differ in scale; a brief table of the relative magnitudes of L_ID, L_N, L_b and L_z after the log-b transform would help readers judge whether re-weighting is warranted.
  2. Figures 5 and 7: the SNR-binned MAE/RMSE panels would be clearer if the number of absorbers per bin were annotated, so that the apparent upturn (or lack thereof) at high SNR can be assessed for Poisson noise.
  3. Section 2.2: the adopted UVB (Faucher-Giguère et al. 2009) and the neglect of non-equilibrium ionization are noted but not quantified; a short statement on how these choices affect the training N_HI distribution relative to UVES would be useful.
  4. Table 1 and architecture description: the Bayesian-optimisation search ranges for window size, filter counts and dense widths are not given; listing them would aid reproducibility.
  5. Throughout: occasional missing spaces after punctuation and inconsistent use of “VIPER” vs “viper” should be cleaned for the final version.

Circularity Check

1 steps flagged

Standard supervised approximation of VIPER labels; evaluation on held-out sims is expected, real-data comparison independent; no derivation reduces to its inputs.

specific steps
  1. other [Section 2.3 / Abstract / Section 4 (sim evaluation)]
    "The model is trained on synthetic spectra generated from the IllustrisTNG simulation and fitted with the VIPER Voigt-profile fitting code to provide training labels. ... On simulated spectra, the CNN achieves an F1 score of ∼0.8 ... It accurately reproduces the H I column density distribution function (CDDF) and the b_HI–N_HI relation, recovering CDDF slopes consistent with VIPER"

    Training and sim-test labels both come from VIPER fits to the identical synthetic spectra; therefore CDDF/b–N agreement on sims is largely the statement that the network successfully approximates its own training distribution. This is expected supervised-ML behaviour rather than a self-definitional or fitted-input prediction of new physics, and the paper’s stronger claim is the (independent) UVES comparison.

full rationale

The paper trains a multi-task CNN to map flux windows to VIPER-derived labels (LyID, log N_HI, log b_HI, z_loc) on TNG synthetic spectra, then reports recovery metrics and CDDF/b–N statistics against the same labeler on held-out sims and against independent VIPER fits on UVES. This is ordinary supervised learning: good sim performance means the network approximates its training distribution, not a circular derivation of a physical quantity. No equation equates a claimed prediction to a fitted input by construction; no uniqueness theorem or ansatz is imported via self-citation to force the result; the latent-space domain-shift analysis and degraded real-data F1/MAE are reported transparently. The central claim (scalable, uncertainty-aware recovery of absorber statistics) therefore stands on independent observational comparison and does not reduce to its training labels. Score 1 only for the minor, non-load-bearing fact that VIPER itself is co-authored by a present author; that does not make the evaluation circular.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central claim rests on the adequacy of TNG+trident mocks labelled by VIPER, a set of architectural and training hyper-parameters chosen by Bayesian optimisation, and the assumption that Monte-Carlo dropout variance is a useful epistemic uncertainty. No new physical entities are introduced; free parameters are the usual ML hyper-parameters and the LyID decision thresholds.

free parameters (4)
  • LyID decision threshold = 0.2 (sim), 0.5 (UVES)
    Chosen post-hoc (0.2 for mocks, 0.5 for UVES) to balance precision/recall; directly controls reported F1 and subsequent statistics.
  • CNN architecture hyper-parameters (filters, kernels, dense widths, dropout, window size) = ws=691, dropout=0.1, filters=(416,224,416), etc.
    Selected by Bayesian optimisation on the validation set; listed in Table 1.
  • Matching velocity threshold Δv_thresh = 10 km s^{-1}
    Set to 10 km s^{-1} for absorber matching; slightly larger than instrumental resolution.
  • N_MC for Monte-Carlo dropout = 50
    Number of stochastic forward passes used to estimate predictive uncertainty.
axioms (4)
  • domain assumption VIPER Voigt-profile fits on TNG sightlines supply ground-truth labels for presence, N_HI, b_HI and centroid.
    Section 2.3; labels are not taken from the simulation particles themselves because of blending.
  • domain assumption trident + uniform UVB + ionisation equilibrium produces sufficiently realistic Lyα forests for training.
    Section 2.2; acknowledged limitations (no local sources, no non-equilibrium).
  • domain assumption Monte-Carlo dropout variance is a useful estimate of epistemic uncertainty.
    Section 3.4, following Gal & Ghahramani (2015).
  • ad hoc to paper Equal weighting of the four task losses is adequate after log-transforming b_HI.
    Section 3.2; chosen after preliminary experiments showed linear b dominated the loss.

pith-pipeline@v1.1.0-grok45 · 26168 in / 2815 out tokens · 19774 ms · 2026-07-11T06:50:40.724043+00:00 · methodology

0 comments
read the original abstract

The Ly$\alpha$ forest is a powerful probe of the intergalactic medium and small-scale matter distribution, but deriving absorber properties traditionally requires computationally expensive Voigt-profile fitting. We present a convolutional neural network (CNN) that identifies and characterizes H I Ly$\alpha$ absorbers directly from quasar spectra. The model is trained on synthetic spectra generated from the IllustrisTNG simulation and fitted with the VIPER Voigt-profile fitting code to provide training labels. The network simultaneously predicts absorber presence, column density ($N_{\rm HI}$), Doppler parameter ($b_{\rm HI}$), and line centroid. On simulated spectra, the CNN achieves an F1 score of $\sim$0.8, with mean absolute errors of $\sim$0.18 in $\log N_{\rm HI}$ and $\sim$0.10 in $\log b_{\rm HI}$. It accurately reproduces the H I column density distribution function (CDDF) and the $b_{\rm HI}$--$N_{\rm HI}$ relation, recovering CDDF slopes consistent with VIPER and a lower-envelope relation with an RMS difference of only 0.36 km s$^{-1}$. Applied to high-resolution UVES spectra, performance decreases to an F1 score of $\sim$0.5, with mean absolute errors of $\sim$0.34 in $\log N_{\rm HI}$ and $\sim$0.21 in $\log b_{\rm HI}$. Latent-space analysis reveals a significant domain shift between the simulated and observational spectra, contributing to the reduced performance. Nevertheless, the CNN preserves the observed CDDF and $b_{\rm HI}$--$N_{\rm HI}$ distributions, yielding CDDF slopes consistent with VIPER and a lower-envelope RMS difference of 2.96 km s$^{-1}$. Monte Carlo dropout is implemented during inference to quantify predictive uncertainties. Together with its computational efficiency, the method provides a scalable and uncertainty-aware framework for Ly$\alpha$ forest analysis in upcoming spectroscopic surveys.

Figures

Figures reproduced from arXiv: 2607.05494 by Hum Chand, Paryag Sharma, Prakash Gaikwad, Ting-Yun Cheng, Vikram Khaire.

Figure 1
Figure 1. Figure 1: Example of a simulated Ly𝛼 forest spectrum generated from the TNG simulation using trident, overlaid with the corresponding Voigt profile fit obtained using viper. 2.3.2 Construction of Training Labels The absorber catalogue generated by viper is used to construct la￾bels for training the CNN, following the general multi-task, sliding￾window framework of Cheng et al. (2022), with several modifications intr… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the sliding-window labeling scheme used in this work. A fixed-size window is scanned across the spectrum with a step of one pixel. Labels are assigned only to the central three pixels around an absorption feature, where Ly𝛼 absorption is present. Pixels within this region are assigned non-zero values for LyID, log 𝑁HI, and log 𝑏HI, while the centroid-offset parameter 𝑧loc takes values 𝑧loc … view at source ↗
Figure 3
Figure 3. Figure 3: Schematic of the one-dimensional convolutional neural network (CNN) used to identify and characterize Ly𝛼 absorption features. The network takes a normalized flux window of size 691 × 1 as input and processes it through three convolutional layers with 416, 224, and 416 filters, respectively. All convolutional layers use a stride of 𝑠 = 1 and are each followed by a max-pooling layer of size 2. The resulting… view at source ↗
Figure 4
Figure 4. Figure 4: Variation of the evaluation metrics as a function of the LyID thresh￾old for simulated spectra. Top panel: Precision (the fraction of predicted absorbers that are truly present) and Recall (the fraction of true absorbers correctly identified by the model), along with their harmonic mean, the F1 score. Bottom panel: Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for the predicted log 𝑁HI and lo… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison between the predicted and true column densities (log 𝑁HI) and Doppler parameter (log 𝑏HI) for simulated spectra. The left panels show the scatter between the predicted and true values of the column density (log 𝑁HI; top) and the Doppler parameter (log 𝑏HI; bottom). In each panel, the solid black line denotes the one-to-one relation (𝑦 = 𝑥). The dashed blue lines mark the loci 𝑦 = 𝑥 ± MAE (mean a… view at source ↗
Figure 6
Figure 6. Figure 6: Same as [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Same as [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Top panel: Comparison of the H i column density distribution function (CDDF) derived from the simulated absorber catalogues identified by the CNN (circles) and viper (squares). Bottom panel: Distribution of Doppler parameter(𝑏HI), as a function of H i column density(log 𝑁HI). The lower-envelope relation is computed using the 10th percentile of 𝑏HI values within logarithmic 𝑁HI bins. Squares denote the vipe… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

76 extracted references · 16 canonical work pages · 16 internal anchors

  1. [1]

    Computational Astrophysics and Cosmology , keywords =

    The IllustrisTNG simulations: public data release. Computational Astrophysics and Cosmology , keywords =. doi:10.1186/s40668-019-0028-x , archivePrefix =. 1812.05609 , primaryClass =

    Pith/arXiv arXiv doi:10.1186/s40668-019-0028-x

  2. [2]

    AREPO: Cosmological magnetohydrodynamical moving-mesh simulation code

  3. [3]

    , keywords =

    Trident: A Universal Tool for Generating Synthetic Absorption Spectra from Astrophysical Simulations. , keywords =. doi:10.3847/1538-4357/aa7e2d , archivePrefix =. 1612.03935 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4357/aa7e2d

  4. [4]

    , keywords =

    VoIgt profile Parameter Estimation Routine (viper): H I photoionization rate at z < 0.5. , keywords =. doi:10.1093/mnras/stx248 , archivePrefix =. 1610.06572 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stx248

  5. [5]

    , keywords =

    Harvesting the Ly forest with convolutional neural networks. , keywords =. doi:10.1093/mnras/stac2631 , archivePrefix =. 2209.02142 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stac2631

  6. [6]

    , keywords =

    Precision Measures of the Primordial Abundance of Deuterium. , keywords =. doi:10.1088/0004-637X/781/1/31 , archivePrefix =. 1308.3240 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/0004-637x/781/1/31

  7. [7]

    Objective measurement of the fine structure constant

    Artificial intelligence applied to the automatic analysis of absorption spectra. Objective measurement of the fine structure constant. , keywords =. doi:10.1093/mnras/stx179 , archivePrefix =. 1606.07393 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stx179

  8. [8]

    arXiv e-prints , keywords =

    An Overview of Multi-Task Learning in Deep Neural Networks. arXiv e-prints , keywords =. doi:10.48550/arXiv.1706.05098 , archivePrefix =. 1706.05098 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1706.05098

  9. [10]

    arXiv e-prints , keywords =

    Adam: A Method for Stochastic Optimization. arXiv e-prints , keywords =. doi:10.48550/arXiv.1412.6980 , archivePrefix =. 1412.6980 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980

  10. [11]

    , keywords =

    The Lyman Alpha Forest in the Spectra of QSOs. , keywords =. doi:10.1146/annurev.astro.36.1.267 , archivePrefix =. astro-ph/9806286 , primaryClass =

    Pith/arXiv arXiv doi:10.1146/annurev.astro.36.1.267

  11. [12]

    , year = 1971, month = mar, volume =

    The Absorption-Line Spectrum of 4c 05.34. , year = 1971, month = mar, volume =. doi:10.1086/180695 , adsurl =

    doi:10.1086/180695 1971

  12. [13]

    , keywords =

    Model-independent Insights into the Nature of the Ly Forest and the Distribution of Matter in the Universe. , keywords =. doi:10.1086/322421 , archivePrefix =. astro-ph/0104272 , primaryClass =

    Pith/arXiv arXiv doi:10.1086/322421

  13. [14]

    , keywords =

    A refined measurement of the mean transmitted flux in the Ly forest over 2 < z < 5 using composite quasar spectra. , keywords =. doi:10.1093/mnras/stt031 , archivePrefix =. 1208.2584 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stt031

  14. [15]

    , keywords =

    The impact of feedback from galaxy formation on the Lyman transmitted flux. , keywords =. doi:10.1093/mnras/sts465 , archivePrefix =. 1207.6567 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/sts465

  15. [16]

    , keywords =

    New constraints on the free-streaming of warm dark matter from intermediate and small scale Lyman- forest data. , keywords =. doi:10.1103/PhysRevD.96.023522 , archivePrefix =. 1702.01764 , primaryClass =

    Pith/arXiv arXiv doi:10.1103/physrevd.96.023522

  16. [17]

    , keywords =

    Strong Bound on Canonical Ultralight Axion Dark Matter from the Lyman-Alpha Forest. , keywords =. doi:10.1103/PhysRevLett.126.071302 , archivePrefix =. 2007.12705 , primaryClass =

    Pith/arXiv arXiv doi:10.1103/physrevlett.126.071302 2007

  17. [18]

    The Space Distribution of the Lyman Alpha Clouds in the Line of Sight to the Z=3.66 QSO 0055-269

    The space distribution of the Lyman alpha clouds in the line of sight to the z=3.66 QSO 0055-269. , keywords =. doi:10.1093/mnras/273.4.1016 , archivePrefix =. astro-ph/9411075 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/273.4.1016

  18. [19]

    The Size and Nature of Lyman alpha Forest Clouds Probed by QSO Pairs and Groups

    The Size and Nature of Lyman- alpha Forest Clouds Probed by QSO Pairs and Groups. , keywords =. doi:10.1086/177129 , archivePrefix =. astro-ph/9510112 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1086/177129

  19. [20]

    , keywords =

    Compressing the Cosmological Information in One-dimensional Correlations of the Lyman- Forest. , keywords =. doi:10.3847/1538-4357/acb433 , archivePrefix =. 2209.09895 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4357/acb433

  20. [21]

    VPFIT: Voigt profile fitting program

  21. [22]

    BayesVP: a Bayesian Voigt profile fitting package

    BayesVP: a Bayesian Voigt profile fitting package. arXiv e-prints , keywords =. doi:10.48550/arXiv.1710.09852 , archivePrefix =. 1710.09852 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1710.09852

  22. [23]

    arXiv e-prints , keywords =

    VoigtFit: A Python package for Voigt profile fitting. arXiv e-prints , keywords =. doi:10.48550/arXiv.1803.01187 , archivePrefix =. 1803.01187 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1803.01187

  23. [24]

    arXiv e-prints , keywords =

    Data Release 1 of the Dark Energy Spectroscopic Instrument. arXiv e-prints , keywords =. doi:10.48550/arXiv.2503.14745 , archivePrefix =. 2503.14745 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.14745

  24. [25]

    Optimising Automatic Morphological Classification of Galaxies with Machine Learning and Deep Learning using Dark Energy Survey Imaging

    Optimizing automatic morphological classification of galaxies with machine learning and deep learning using Dark Energy Survey imaging. , keywords =. doi:10.1093/mnras/staa501 , archivePrefix =. 1908.03610 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/staa501 1908

  25. [26]

    Galaxy Morphological Classification Catalogue of the Dark Energy Survey Year 3 data with Convolutional Neural Networks

    Galaxy morphological classification catalogue of the Dark Energy Survey Year 3 data with convolutional neural networks. , keywords =. doi:10.1093/mnras/stab2142 , archivePrefix =. 2107.10210 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stab2142

  26. [27]

    , keywords =

    Galaxy Zoo DECaLS: Detailed visual morphology measurements from volunteers and deep learning for 314 000 galaxies. , keywords =. doi:10.1093/mnras/stab2093 , archivePrefix =. 2102.08414 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stab2093

  27. [28]

    The Strong Gravitational Lens Finding Challenge

    The strong gravitational lens finding challenge. , keywords =. doi:10.1051/0004-6361/201832797 , archivePrefix =. 1802.03609 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004-6361/201832797

  28. [29]

    RAPID: Early Classification of Explosive Transients using Deep Learning

    RAPID: Early Classification of Explosive Transients Using Deep Learning. , keywords =. doi:10.1088/1538-3873/ab1609 , archivePrefix =. 1904.00014 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/1538-3873/ab1609 1904

  29. [30]

    , keywords =

    Detecting damped Ly absorbers with Gaussian processes. , keywords =. doi:10.1093/mnras/stx1958 , archivePrefix =. 1605.04460 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stx1958

  30. [31]

    , keywords =

    Deep learning of quasar spectra to discover and characterize damped Ly systems. , keywords =. doi:10.1093/mnras/sty196 , archivePrefix =. 1709.04962 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/sty196

  31. [32]

    Deep Forest: Neural Network reconstruction of the Lyman-alpha forest

    Deep forest: Neural network reconstruction of the Lyman- forest. , keywords =. doi:10.1093/mnras/stab2041 , archivePrefix =. 2009.10673 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stab2041 2009

  32. [33]

    Reconstruction of the Density Power Spectrum from Quasar Spectra using Machine Learning

    Reconstruction of the Density Power Spectrum from Quasar Spectra using Machine Learning. arXiv e-prints , keywords =. doi:10.48550/arXiv.2107.09082 , archivePrefix =. 2107.09082 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2107.09082

  33. [34]

    Deep Learning Voigt Profiles. I. Single-Cloud Doublets. , keywords =. doi:10.3847/1538-3881/ad402b , archivePrefix =. 2311.16029 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-3881/ad402b

  34. [35]

    , keywords =

    Automated quasar continuum estimation using neural networks: A comparative study of deep-learning architectures. , keywords =. doi:10.1051/0004-6361/202453377 , archivePrefix =. 2505.10976 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/202453377

  35. [36]

    Application of Convolutional Neural Networks for Stellar Spectral Classification

    Application of convolutional neural networks for stellar spectral classification. , keywords =. doi:10.1093/mnras/stz3100 , archivePrefix =. 1909.05459 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stz3100 1909

  36. [37]

    , keywords =

    A neural network emulator for the Lyman- forest 1D flux power spectrum. , keywords =. doi:10.1093/mnras/stad2512 , archivePrefix =. 2305.19064 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stad2512

  37. [38]

    , keywords =

    Towards a unified description of the intergalactic medium at redshift z 2.5. , keywords =. doi:10.1093/mnras/stt2218 , archivePrefix =. 1310.0052 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stt2218

  38. [39]

    The b Distribution and the Velocity Structure of Absorption Peaks in the Lyman-Alpha Forest

    The b Distribution and the Velocity Structure of Absorption Peaks in the Ly Forest. , keywords =. doi:10.1086/307202 , archivePrefix =. astro-ph/9709100 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1086/307202

  39. [40]

    , keywords =

    FLAME: Fitting Ly absorption lines using machine learning. , keywords =. doi:10.1051/0004-6361/202449756 , archivePrefix =. 2403.07498 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/202449756

  40. [41]

    An HST/COS Survey of the Low-redshift Intergalactic Medium. I. Survey, Methodology, and Overall Results. , keywords =. doi:10.3847/0004-637X/817/2/111 , archivePrefix =. 1402.2655 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/0004-637x/817/2/111

  41. [42]

    , keywords =

    The UVES Spectral Quasar Absorption Database (SQUAD) data release 1: the first 10 million seconds. , keywords =. doi:10.1093/mnras/sty2834 , archivePrefix =. 1810.06136 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/sty2834

  42. [43]

    , keywords =

    The AREPO Public Code Release. , keywords =. doi:10.3847/1538-4365/ab908c , archivePrefix =. 1909.04667 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4365/ab908c 1909

  43. [44]

    , keywords =

    A cosmic UV/X-ray background model update. , keywords =. doi:10.1093/mnras/staa302 , archivePrefix =. 1903.08657 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/staa302 1903

  44. [45]

    , keywords =

    Simulating galaxy formation with the IllustrisTNG model. , keywords =. doi:10.1093/mnras/stx2656 , archivePrefix =. 1703.02970 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stx2656

  45. [46]

    , keywords =

    The 2017 Release Cloudy. , keywords =. doi:10.48550/arXiv.1705.10877 , archivePrefix =. 1705.10877 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1705.10877 2017

  46. [47]

    arXiv e-prints , keywords =

    Deep Learning using Rectified Linear Units (ReLU). arXiv e-prints , keywords =. doi:10.48550/arXiv.1803.08375 , archivePrefix =. 1803.08375 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1803.08375

  47. [48]

    arXiv e-prints , keywords =

    Practical Bayesian Optimization of Machine Learning Algorithms. arXiv e-prints , keywords =. doi:10.48550/arXiv.1206.2944 , archivePrefix =. 1206.2944 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1206.2944

  48. [49]

    The physics of the intergalactic medium , author =. Rev. Mod. Phys. , volume =. 2009 , month =. doi:10.1103/RevModPhys.81.1405 , url =

    doi:10.1103/revmodphys.81.1405 2009

  49. [50]

    , keywords =

    Baryon acoustic oscillations in the Ly forest of BOSS quasars. , keywords =. doi:10.1051/0004-6361/201220724 , archivePrefix =. 1211.2616 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/201220724

  50. [51]

    , keywords =

    Measurement of baryon acoustic oscillations in the Lyman- forest fluctuations in BOSS data release 9. , keywords =. doi:10.1088/1475-7516/2013/04/026 , archivePrefix =. 1301.3459 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/1475-7516/2013/04/026 2013

  51. [52]

    , keywords =

    The Completed SDSS-IV Extended Baryon Oscillation Spectroscopic Survey: Baryon Acoustic Oscillations with Ly Forests. , keywords =. doi:10.3847/1538-4357/abb085 , archivePrefix =. 2007.08995 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4357/abb085 2007

  52. [53]

    , keywords =

    A New Calculation of the Ionizing Background Spectrum and the Effects of He II Reionization. , keywords =. doi:10.1088/0004-637X/703/2/1416 , archivePrefix =. 0901.4554 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/0004-637x/703/2/1416

  53. [54]

    Radiative Transfer in a Clumpy Universe. IV. New Synthesis Models of the Cosmic UV/X-Ray Background. , keywords =. doi:10.1088/0004-637X/746/2/125 , adsurl =

    doi:10.1088/0004-637x/746/2/125

  54. [55]

    , keywords =

    The EAGLE simulations of galaxy formation: calibration of subgrid physics and model variations. , keywords =. doi:10.1093/mnras/stv725 , archivePrefix =. 1501.01311 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stv725

  55. [56]

    , keywords =

    SIMBA: Cosmological simulations with black hole growth and feedback. , keywords =. doi:10.1093/mnras/stz937 , archivePrefix =. 1901.10203 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stz937 1901

  56. [57]

    , keywords =

    Nyx: A Massively Parallel AMR Code for Computational Cosmology. , keywords =. doi:10.1088/0004-637X/765/1/39 , archivePrefix =. 1301.4498 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/0004-637x/765/1/39

  57. [58]

    , keywords =

    The Sherwood simulation suite: overview and data comparisons with the Lyman forest at redshifts 2 z 5. , keywords =. doi:10.1093/mnras/stw2397 , archivePrefix =. 1605.03462 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stw2397

  58. [59]

    Can the Low Redshift Lyman Alpha Forest Constrain AGN Feedback Models?

    Can the low-redshift Lyman alpha forest constrain AGN feedback models?. , keywords =. doi:10.1093/mnras/stad3374 , archivePrefix =. 2306.05466 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stad3374

  59. [60]

    The Impact of the WHIM on the IGM Thermal State Determined from the Low-$z$ Lyman-$\alpha$ Forest

    The impact of the WHIM on the IGM thermal state determined from the low-z Lyman forest. , keywords =. doi:10.1093/mnras/stad3846 , archivePrefix =. 2308.14738 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stad3846

  60. [61]

    arXiv e-prints , keywords =

    Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning. arXiv e-prints , keywords =. doi:10.48550/arXiv.1506.02142 , archivePrefix =. 1506.02142 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1506.02142

  61. [62]

    , keywords =

    The evolution of the low-density H I> intergalactic medium from z = 3.6 to 0: data, transmitted flux, and H I> column density,. , keywords =. doi:10.1093/mnras/staa3844 , archivePrefix =. 2012.05861 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/staa3844 2012

  62. [63]

    , keywords =

    The Temperature-Density Relation in the Intergalactic Medium at Redshift langzrang = 2.4. , keywords =. doi:10.1088/2041-8205/757/2/L30 , archivePrefix =. 1209.0005 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/2041-8205/757/2/l30 2041

  63. [64]

    , keywords =

    Intergalactic Lyman continuum photon budget in the past 5 billion years. , keywords =. doi:10.1093/mnras/stw3086 , archivePrefix =. 1605.02738 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stw3086

  64. [65]

    , keywords =

    Probing the thermal state of the intergalactic medium at z > 5 with the transmission spikes in high-resolution Ly forest spectra. , keywords =. doi:10.1093/mnras/staa907 , archivePrefix =. 2001.10018 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/staa907 2001

  65. [66]

    , keywords =

    A consistent and robust measurement of the thermal state of the IGM at 2 z 4 from a large sample of Ly forest spectra: evidence for late and rapid He II reionization. , keywords =. doi:10.1093/mnras/stab2017 , archivePrefix =. 2009.00016 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stab2017 2009

  66. [67]

    A Measurement of the Thermal and Ionization State of the IGM at $z < 0.5$

    A Measurement of the Thermal and Ionization State of the IGM at z < 0.5. arXiv e-prints , keywords =. doi:10.48550/arXiv.2606.05006 , archivePrefix =. 2606.05006 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2606.05006

  67. [68]

    , keywords =

    Possible evidence for an inverted temperature-density relation in the intergalactic medium from the flux distribution of the Ly forest. , keywords =. doi:10.1111/j.1365-2966.2008.13114.x , archivePrefix =. 0711.2064 , primaryClass =

    doi:10.1111/j.1365-2966.2008.13114.x 2008

  68. [69]

    , keywords =

    The Lyman forest opacity and the metagalactic hydrogen ionization rate at z -0.5ex 2-4. , keywords =. doi:10.1111/j.1365-2966.2005.08704.x , archivePrefix =. astro-ph/0411072 , primaryClass =

    doi:10.1111/j.1365-2966.2005.08704.x 2005

  69. [70]

    , keywords =

    New measurements of the ionizing ultraviolet background over 2 < z < 5 and implications for hydrogen reionization. , keywords =. doi:10.1093/mnras/stt1610 , archivePrefix =. 1307.2259 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stt1610

  70. [71]

    , keywords =

    Measurements of the ultraviolet background at 4.6 < z < 6.4 using the quasar proximity effect. , keywords =. doi:10.1111/j.1365-2966.2010.18072.x , archivePrefix =. 1011.5850 , primaryClass =

    doi:10.1111/j.1365-2966.2010.18072.x 2010

  71. [72]

    , keywords =

    New synthesis models of consistent extragalactic background light over cosmic time. , keywords =. doi:10.1093/mnras/stz174 , archivePrefix =. 1801.09693 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stz174

  72. [73]

    Star formation history, dust attenuation and extragalactic background light

    Star Formation History, Dust Attenuation, and Extragalactic Background Light. , keywords =. doi:10.1088/0004-637X/805/1/33 , archivePrefix =. 1405.7038 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/0004-637x/805/1/33

  73. [74]

    , keywords =

    Consistent modelling of the meta-galactic UV background and the thermal/ionization history of the intergalactic medium. , keywords =. doi:10.1093/mnras/stz222 , archivePrefix =. 1801.04931 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stz222

  74. [75]

    , keywords =

    Measurements of the thermal and ionization state of the intergalactic medium during the cosmic afternoon. , keywords =. doi:10.1093/mnras/stae2474 , archivePrefix =. 2311.17895 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stae2474

  75. [76]

    , keywords =

    Completed SDSS-IV extended Baryon Oscillation Spectroscopic Survey: Cosmological implications from two decades of spectroscopic surveys at the Apache Point Observatory. , keywords =. doi:10.1103/PhysRevD.103.083533 , archivePrefix =. 2007.08991 , primaryClass =

    Pith/arXiv arXiv doi:10.1103/physrevd.103.083533 2007

  76. [77]

    , keywords =

    The power spectrum of the Lyman- Forest at z < 0.5. , keywords =. doi:10.1093/mnras/stz344 , archivePrefix =. 1808.05605 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stz344