Pith. sign in

REVIEW 4 major objections 6 minor 117 references

TPCNet: Representation learning for HI mapping

T0 review · 4 major / 6 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 hybrid CNN-Transformer network can infer cold gas fraction and opacity correction from HI emission spectra alone, beating deep CNNs by about 10% and matching absorption-based estimates in the optically thin regime.

desk verdict Useful machine-learning paper with an honest but unresolved transferability gap: the synthetic training set lacks H2 formation, so the headline agreement only holds in optically-thin regimes. read the letter →

arxiv 2411.13325 v1 pith:HDIXAUQX submitted 2024-11-20 astro-ph.GA

Hiep Nguyen , Haiyang Tang , Matthew Alger , Antoine Marchal , Eric G. M. Muller , Cheng Soon Ong , N. M. McClure-Griffiths This is my paper · ORCID
classification astro-ph.GA
keywords 21-cmemissioncoldneutralmediumHIopacitycorrectionconvolutionalneuralnetworkTransformerpositionalencodingsyntheticspectralcubesGalacticISM
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

TPCNet, a hybrid convolutional-Transformer network, is claimed to learn the cold neutral medium fraction $f_{\rm CNM}$ and the HI opacity correction factor $R_{\rm HI}$ directly from 21-cm emission spectra, without the absorption measurements normally required. Trained on synthetic spectra from hydrodynamic and magnetohydrodynamic simulations, it reportedly outperforms deep CNNs by about 10% in testing accuracy, training stability, and convergence speed. On observed data, TPCNet predictions agree with Gaussian-decomposition-based estimates in the optically thin regime ($N_{\rm HI}^* \lesssim 5\times10^{20}$ cm$^{-2}$), but run higher than absorption-based values toward the Galactic plane. The paper's central claim is that emission-only large-area cold-gas mapping is feasible.

What carries the argument

The load-bearing object is the TPCNet architecture: an 8-layer convolutional network with iterated kernel sizes 7 and 33 (velocity channels) that reduces the spectrum to a feature vector, which is reshaped into a $m \times k$ token embedding and passed to a Transformer decoder. Multi-head self-attention computes attention scores across tokens, letting the model associate spectral features at widely separated velocities. The sinusoidal positional encoding added to the input spectrum supplies the order information that lets the model handle spectra whose signal is not centered in the velocity window, a known failure mode of the earlier shallow CNN.

What would settle it

Retrain TPCNet with identical architecture on synthetic cubes from a simulation that includes the atomic-to-molecular transition, then compare its predictions on the 157 BIGHICAT absorption sightlines. If the overprediction at $N_{\rm HI}^* > 5\times10^{20}$ cm$^{-2}$ (up to $\Delta f_{\rm CNM} \sim 0.7$) does not shrink, the discrepancy is not caused by the missing transition; if it does shrink, the current models' high-column-density bias is confirmed as a training-set artifact.

Watch

Extended reading notes

Core claim

In the paper's own terms, TPCNet extracts compact features from HI emission spectra with an 8-layer CNN and feeds them to a Transformer decoder, whose multi-head self-attention captures long-range correlations between velocity channels. The selected 'add sinusoidal' positional encoding, adding a sinusoidal function of channel index directly to the spectrum, makes predictions robust to dataset shuffling and convolutional weight initialization. On the synthetic evaluation cube, the model reproduces $f_{\rm CNM}$ and $R_{\rm HI}$ ground truths with RMSE of 3.5% and 0.05 respectively. On 157 absorption lines of sight, predictions match the absorption-based values for $N_{\rm HI}^* < 5\times10^{20}$ cm$^{-2}$, while $\Delta f_{\rm CNM}$ reaches up to ~0.7 at higher column densities; the paper attributes this deviation partly to the simulations' lack of an atomic-to-molecular transition and partly to possible underfitting in the emission-absorption Gaussian decomposition.

Load-bearing premise

The load-bearing premise is that synthetic spectra from simulations matching Solar-neighborhood conditions, and lacking molecular gas formation, are representative enough of real Galactic sightlines for the learned emission-to-cold-gas mapping to transfer.

Editorial extensions

If this is right

  • Emission-only HI surveys could produce wide-area maps of $f_{\rm CNM}$ and $R_{\rm HI}$, bypassing the sparse continuum sources needed for absorption measurements.
  • Higher spectral resolution (0.3125 km s$^{-1}$, 256 channels) improves prediction accuracy over 0.8 km s$^{-1}$, favoring high-resolution surveys for this technique.
  • Training on synthetic cubes from multiple simulations generalizes better than training on a single simulation, suggesting that diversity of the training set is a controllable lever for accuracy.
  • The same hybrid architecture is proposed for other spectral lines, such as CO, to infer molecular gas properties from emission alone.

Reading between the lines

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

  • Because the simulations omit the atomic-to-molecular transition, the high-column-density overprediction is plausibly a synthetic training bias; retraining on a simulation that includes H$_2$ formation would test this directly.
  • The bounded range of $f_{\rm CNM}$ (0 to 1) makes it an easier regression target than $R_{\rm HI}$ (unbounded above), which may explain the larger scatter in $R_{\rm HI}$ predictions; a bounded or probabilistic output layer could improve the high-$R_{\rm HI}$ tail.
  • The positional encoding's ability to handle off-center spectral peaks suggests TPCNet could be applied to intermediate-velocity and high-latitude clouds without spectral recentering, a practical advantage over the M20 CNN.
  • If the learned emission-to-opacity mapping transfers, the method could be extended to estimate spin temperature distributions or optical depth directly from emission, which the current work does not attempt.
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

4 major / 6 minor

Summary. The paper introduces TPCNet, a hybrid CNN-Transformer network with sinusoidal positional encoding, trained on synthetic HI spectra from two numerical simulations (Saury et al. 2014; Seta & Federrath 2022) to predict the cold neutral medium fraction f_CNM and the HI opacity correction factor R_HI from 21-cm emission spectra alone. The synthetic evaluation shows low RMSE (3.5% for f_CNM and 0.05 for R_HI) on a held-out cube from the same simulation suite. Applications to observed absorption survey sightlines and to the LLIV1 emission cube are compared with Gaussian-decomposition-based estimates. The paper claims a 10% average improvement over deep CNNs in testing accuracy, stability, and convergence speed, and strong agreement with observed estimates.

Significance. If the transferability claim holds, the method would enable large-area mapping of cold HI from emission-only data, avoiding the need for rare absorption pairs. The architecture is novel in this context, with a careful ablation of positional encodings, kernel sizes, spectral resolutions, and noise levels. The code and data are publicly available, which is a strength. However, the synthetic training database is acknowledged to lack the atomic-to-molecular transition, leading to an excess of cold optically-thick HI. The observed-data validation shows systematic overpredictions in the high-column-density regime, so the headline claim of emission-only mapping is not yet established.

major comments (4)
  1. [Section 2.1 and Figure 2] The synthetic training distribution is biased relative to observations: synthetic f_CNM extends to 100% and R_HI to ~15, whereas absorption-based values reach only 88% and ~3. The paper explicitly states that the simulations lack the atomic-to-molecular transition, which leaves an excess of cold optically-thick HI in the training cubes. Consequently, the synthetic evaluation RMSEs reported in Section 4 (3.5% for f_CNM, 0.05 for R_HI) measure in-distribution fit to this biased distribution, not physical accuracy. The claim that TPCNet generalizes to observed data is therefore not supported in the high-column-density regime where the training distribution is most unrepresentative.
  2. [Section 5, Figure 9] At N_HI* > 5e20 cm^-2, TPCNet overpredicts f_CNM by up to ~0.7 relative to absorption-based estimates from 21-SPONGE and Millennium surveys. This is exactly the regime where opacity corrections are most important, and the paper itself attributes the discrepancy to the missing atomic-to-molecular transition. The paper offers a second possible explanation (underestimation by Gaussian fitting) but does not provide any test to distinguish between these two hypotheses. Without such a test, the observed-data validation cannot support the abstract's claim of 'strong agreement' beyond the optically-thin regime.
  3. [Section 6.3, Figure 13] The ROHSA comparison reaches its best correlation (R=0.92) only after subtracting all CNM Gaussian components with amplitudes T_B < 2 K from the reference map, which drops the RMSE from 0.138 to 0.088. This subtraction is not justified physically; it removes real emission that ROHSA identifies as CNM. The paper thus claims agreement after post-hoc modification of the reference data, which undermines the validation on the observed emission cube. The abstract's statement of 'strong agreement between the predictions and Gaussian decomposition-based estimates' is therefore not supported by the ROHSA comparison as presented.
  4. [Abstract and Section 7] The abstract and conclusions overstate the findings. The abstract claims 'strong agreement' with Gaussian decomposition-based estimates, but the body of the paper shows large deviations in the high-column-density regime (Section 5) and only conditional agreement with ROHSA after the subtraction described above. The '10% average increase in testing accuracy' is measured on synthetic spectra and may not translate to real data, where the model's performance is limited by the training bias. The paper should temper these claims or provide additional validation that the mapping transfers to the real ISM.
minor comments (6)
  1. [Throughout] The notation 'Hi' is non-standard; the chemical symbol for neutral atomic hydrogen is 'H I' (or 'HI' as an abbreviation). The manuscript should use a consistent form.
  2. [Figure 6 caption] The caption states 'with a mean (standard deviation) of 12% (21%) for f_CNM and 1% (3%) for f_CNM'; the second variable should be R_HI.
  3. [Section 4] The text says the relative difference 'ranges from -250% to 92%' but then says values below -92% are clipped; this is inconsistent and should be clarified.
  4. [Section 2.3] The definition of CNM uses T_k < 500 K in Section 2.1, but Section 2.3 refers to 'T_s < 500 K'. Please reconcile the temperature notation.
  5. [Section D2] The phrase 'the RMSE error' is redundant; use simply 'RMSE'.
  6. [Section 3 and Appendix D] The paper uses 'testing accuracy' while the evaluation metric is RMSE; 'accuracy' typically refers to classification. Consider using 'prediction error' or 'RMSE' for clarity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TPCNet is a supervised regressor trained on synthetic spectra and tested on independent held-out and observed benchmarks; the acknowledged simulation-vs-observation mismatch is a transferability caveat, not a circular reduction.

full rationale

The paper's central pipeline is standard supervised regression: synthetic spectra with known f_CNM and R_HI labels (computed from radiative transfer of simulated density, temperature, and velocity fields) are used to train TPCNet, and the model is then evaluated on a held-out synthetic cube and on observed emission data compared with absorption surveys, ROHSA, and Fourier-transform results. None of these evaluations uses the target observed f_CNM or R_HI values as training inputs, and no equation in the paper defines the predicted quantities in terms of the model outputs or vice versa. The 'learned relationships' in Section 2.2 are descriptive trends in the training database, not a derivation of the output from the input. The held-out evaluation RMSE (f_CNM 3.5%, R_HI 0.05) is a standard generalization test, not a tautology. The observed-data comparisons are external: BIGHICAT absorption measurements, ROHSA decompositions, and Marchal et al. (2024) Fourier transforms are independent of TPCNet's fitted parameters. The acknowledged limitation that the simulations 'do not include the atomic-to-molecular transition' and produce an excess of cold optically-thick HI is a domain-shift / transferability concern, explicitly stated by the authors, and it is not a circularity. Similarly, the ROHSA comparison's post-hoc subtraction of small-amplitude CNM components is a reference-adjustment caveat that affects the strength of the agreement, but it does not make the model's prediction equivalent to its input by construction. Self-citations to Marchal et al. (2019, 2024) and McClure-Griffiths et al. (2023) provide tools, baselines, and catalogs; they are not used as an unverified uniqueness theorem or as the sole justification of the core claim. Therefore the derivation chain is not circular.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the representativeness of two simulations for the real ISM, the 21-cm radiative transfer used to create labels, and the reliability of Gaussian-decomposition reference values. The model adds many fitted parameters (network weights) plus hand-tuned hyperparameters (8 layers, kernels 7 and 33, learning rates, token length k=9 explicitly 'arbitrary'). No new physical entities are introduced.

free parameters (6)
  • Neural network weights and biases = fitted to 1,048,576 synthetic spectra
    The TPCNet and deep CNN weights are fitted to the synthetic training data; the learned emission-to-fraction mapping is the core of the prediction, and whether it transfers to the real ISM is the main uncertainty.
  • Learning rate = 5e-3 (256 ch); 5e-2 CNN / 9e-3 TPCNet (101 ch)
    Chosen by experiment (Appendix B1); material to convergence and final RMSE.
  • Number of convolutional layers = 8
    Selected from 2 to 14 layer scans (Figure B1) as the trade-off between underfitting and overfitting.
  • Convolutional kernel sizes = 7 and 33 channels
    Iteratively used to capture CNM- and WNM-scale features; a kernel-size scan (Figure D5) shows an RMSE minimum near 39 channels, so the chosen sizes are justified but tuned.
  • Transformer embedding token length k = 9
    Explicitly described as 'arbitrary' in Appendix A3; the paper reports RMSE is insensitive to the choice.
  • Synthetic noise level and beam size = 0.5-0.8 K noise; 45 arcsec beam
    Chosen to mimic real observations; Appendix D3 notes HPC limits prevented a systematic scan of observing parameters.
assumptions (5)
  • standard math The 21-cm radiative transfer relations (Equations 1 and 2) connect brightness temperature, optical depth, spin temperature, and column density.
    Standard radio astronomy; used to generate synthetic training labels and to convert brightness to optically-thin column density N_HI*.
  • domain assumption The HD/MHD simulations of Saury et al. (2014) and Seta and Federrath (2022) reproduce realistic multiphase ISM conditions so that mappings learned on synthetic spectra transfer to the Galaxy.
    The paper acknowledges in Section 2.1 that the simulations match only Solar-neighborhood parameters and lack the atomic-to-molecular transition, leaving excess cold optically-thick HI; this is the source of the high-column-density overprediction seen in Section 5.
  • domain assumption Heiles and Troland (2003a) style Gaussian decomposition of emission-absorption pairs gives accurate reference f_CNM and R_HI values.
    Used as the validation ground truth in Section 5; the paper itself notes (Section 5) that Gaussian fitting lacks uniqueness and can underestimate CNM components.
  • domain assumption A spectral resolution of 0.3125 km/s (256 channels) resolves CNM spectral features; regridding cannot recover information lost at lower resolution.
    Underpins the resolution study in Appendix D3 showing that 256-channel training beats 101-channel training and that regridding between resolutions does not help.
  • domain assumption CNM is defined as HI gas with kinetic temperature T_k < 500 K (Heiles and Troland 2003a convention).
    Defines the target f_CNM and the recomputation of absorption-survey values; the ROHSA comparison uses a looser 1000 K threshold, complicating the comparison (Section 6.3).

how reviews work

0 comments
Cite this review

Pith. "Pith review of TPCNet: Representation learning for HI mapping." pith.science (2026). https://pith.science/paper/HDIXAUQX

@misc{pith2026241113325,
  author       = {Pith},
  title        = {Pith review of: TPCNet: Representation learning for HI mapping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HDIXAUQX}},
  note         = {Machine review of arXiv:2411.13325}
}
abstract

We introduce TPCNet, a neural network predictor that combines Convolutional and Transformer architectures with Positional encodings, for neutral atomic hydrogen (HI) spectral analysis. Trained on synthetic datasets, our models predict cold neutral gas fraction ($f_\text{CNM}$) and HI opacity correction factor ($R_\text{HI}$) from emission spectra based on the learned relationships between the desired output parameters and observables (optically-thin column density and peak brightness). As a follow-up to Murray et al. (2020)'s shallow Convolutional Neural Network (CNN), we construct deep CNN models and compare them to TPCNet models. TPCNet outperforms deep CNNs, achieving a 10% average increase in testing accuracy, algorithmic (training) stability, and convergence speed. Our findings highlight the robustness of the proposed model with sinusoidal positional encoding applied directly to the spectral input, addressing perturbations in training dataset shuffling and convolutional network weight initializations. Higher spectral resolutions with increased spectral channels offer advantages, albeit with increased training time. Diverse synthetic datasets enhance model performance and generalization, as demonstrated by producing $f_\text{CNM}$ and $R_\text{HI}$ values consistent with evaluation ground truths. Applications of TPCNet to observed emission data reveal strong agreement between the predictions and Gaussian decomposition-based estimates (from emission and absorption surveys), emphasizing its potential in HI spectral analysis.

Figures

Figures reproduced from arXiv: 2411.13325 by the authors.

Figure 1
Figure 1. Samples of synthetic Hi spectra in training database derived from Seta & Federrath (2022) (left panels) and Saury et al. (2014) (right panels), each paired with their associated CNM fraction 𝑓CNM (with 𝑇k, CNM < 500 K) and opacity correction to optically-thin estimate RHI. The velocity channel bin width is 0.3125 km s−1 . optical depth can be directly measured only in the directions of strong radio continuum sources… view at source ↗
Figure 3
Figure 3. Hi maps computed from a training spectral cube generated by Saury et al. (2014)’s simulation with a resolution (cell size) of 0.04 pc: Optically￾thin Hi column density 𝑁∗ HI (upper left), peak brightness temperature 𝑇b,peak (upper right), cold gas mass fraction 𝑓CNM with 𝑇k, CNM < 500 K (lower left), and Hi opacity correction factor RHI (lower right). absorption observations (magenta lines), compiled by McClure￾Grif… view at source ↗
Figure 4
Figure 4. The relationships between ground truth (“gt”) values of CNM mass fraction 𝑓CNM, opacity correction factor RHI and optically-thin Hi column density 𝑁∗ HI (first column), maximum brightness temperature 𝑇b,peak (second column), equivalent width EW = ∫ 𝜏𝑑𝑣 [km s−1 ] (third column), and Hi peak optical depth 𝜏peak (last column) in a synthetic position-position-velocity spectral data cube [PITH_FULL_IMAGE:figures/full_fi… view at source ↗
Figures from the paper (11 more)
Figure 5
Figure 5. Figure 5: The relationships between ground truth (“gt”) values of CNM mass fraction 𝑓CNM and opacity correction factor RHI in the same synthetic position-position-velocity spectral data cube as in [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Predictions of 𝑓CNM (two first columns) and RHI (two last columns) by TPCNets using the evaluation cube with a cell size of 0.04 pc. Upper left: ground-truth (“gt”); upper right: predictions (“pred”); lower left: relative difference between ground truths and prediction…
Figure 7
Figure 7. Figure 7: Same as [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Locations of absorption measurements from “BIGHICAT” catalog (McClure-Griffiths et al. 2023) in the Galactic coordinates, overlaid on the map of HI4PI column density 𝑁∗ HI (HI4PI Collaboration et al. 2016). The “X” marker (at 𝑙 = 0 ◦ , 𝑏 = 0 ◦ ) labels the Galactic cen…
Figure 9
Figure 9. Figure 9: Comparison between absorption cold Hi gas fraction ( 𝑓CNM,abs) and TPCNet predictions ( 𝑓CNM,pred) using observed data from absorption surveys: stars for 21-SPONGE, circles for MACH, and triangles for Millennium Survey. Color scale indicates the absolute Galactic latit…
Figure 10
Figure 10. Figure 10: Same as [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Predictions by representation learning TPCNet model for the Low-Latitude Intermediate-Velocity Arch 1 (LLIV1) observed by the GHIGLS survey: Maps of peak brightness temperature 𝑇b,peak (top left) and optically-thin column density 𝑁∗ HI (bottom left, in the unit of 102…
Figure 12
Figure 12. Figure 12: Comparison between TPCNet predictions and Fourier-transformed method for the Low-Latitude Intermediate-Velocity Arch 1: CNM column density estimated by Fourier transform (top left) and by TPCNet model (top right) both in unit of 1020 cm−2 , along with their one-to-one…
Figure 14
Figure 14. Figure 14: The relationships between 𝑓CNM, RHI and 𝑁∗ HI (left column), 𝑇b,peak (right column) in the LLIV1 observed dataset. Top row for 𝑓CNM obtained from ROHSA Gaussian decomposition; middle row for 𝑓CNM pre￾dicted by TPCNet model. The y-axis scales in the first two rows are …
Figure 16
Figure 16. Figure 16: Comparison between TPCNet predictions and M20 CNN for the LLIV1: CNM column density 𝑁HI,CNM estimated by TPCNet model (top left) and by M20 CNN (top right) both in unit of 1020 cm−2 , along with their one-to-one comparison (bottom left) and the dependence of their abs…
Figure 17
Figure 17. Figure 17: Comparison between TPCNet predictions and M20 CNN for the HI4PI IVC at high Galactic latitude, centered at (𝑙, 𝑏) = (+135◦ , +55◦ ): CNM column density 𝑁HI,CNM estimated by TPCNet model (top left) and by M20 CNN (top right) both in unit of 1020 cm−2 , along with their…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

117 extracted references · 15 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTION or pop #1...

  2. [2]

    Abdel-Hamid O., Mohamed A.-r., Jiang H., Deng L., Penn G., Yu D., 2014, @doi [IEEE/ACM Transactions on Audio, Speech, and Language Processing] 10.1109/TASLP.2014.2339736 , 22, 1533

    arXiv 2014

  3. [3]

    Astropy Collaboration et al., 2018, @doi [ ] 10.3847/1538-3881/aabc4f , https://ui.adsabs.harvard.edu/abs/2018AJ....156..123A 156, 123

    doi:10.3847/1538-3881/aabc4f 2018

  4. [4]

    Audit E., Hennebelle P., 2005, @doi [ ] 10.1051/0004-6361:20041474 , http://adsabs.harvard.edu/abs/2005A

    doi:10.1051/0004-6361:20041474 2005

  5. [5]

    Audit E., Hennebelle P., 2010, @doi [ ] 10.1051/0004-6361/200912695 , https://ui.adsabs.harvard.edu/abs/2010A&A...511A..76A 511, A76

    doi:10.1051/0004-6361/200912695 2010

  6. [6]

    V., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1904.09925 , https://ui.adsabs.harvard.edu/abs/2019arXiv190409925B p

    Bello I., Zoph B., Vaswani A., Shlens J., Le Q. V., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1904.09925 , https://ui.adsabs.harvard.edu/abs/2019arXiv190409925B p. arXiv:1904.09925

    arXiv doi:10.48550/arxiv.1904.09925 2019

  7. [7]

    arXiv:1206.5538

    Bengio Y., Courville A., Vincent P., 2012, @doi [arXiv e-prints] 10.48550/arXiv.1206.5538 , https://ui.adsabs.harvard.edu/abs/2012arXiv1206.5538B p. arXiv:1206.5538

    arXiv doi:10.48550/arxiv.1206.5538 2012

  8. [8]

    arXiv:2303.12712

    Bubeck S., et al., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2303.12712 , https://ui.adsabs.harvard.edu/abs/2023arXiv230312712B p. arXiv:2303.12712

    arXiv doi:10.48550/arxiv.2303.12712 2023

Show all 117 references
  1. [9]

    arXiv:2005.12872

    Carion N., Massa F., Synnaeve G., Usunier N., Kirillov A., Zagoruyko S., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2005.12872 , https://ui.adsabs.harvard.edu/abs/2020arXiv200512872C p. arXiv:2005.12872

    2020 doi arXiv

  2. [10]

    arXiv:2006.09882

    Caron M., Misra I., Mairal J., Goyal P., Bojanowski P., Joulin A., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2006.09882 , https://ui.adsabs.harvard.edu/abs/2020arXiv200609882C p. arXiv:2006.09882

    2020 doi arXiv

  3. [11]

    A., et al., 2015, @doi [ ] 10.1093/mnras/stv725 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.450.1937C 450, 1937

    Crain R. A., et al., 2015, @doi [ ] 10.1093/mnras/stv725 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.450.1937C 450, 1937

    2015 doi

  4. [12]

    H., Appleby S., 2019, @doi [ ] 10.1093/mnras/stz937 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.486.2827D 486, 2827

    Dav \'e R., Angl \'e s-Alc \'a zar D., Narayanan D., Li Q., Rafieferantsoa M. H., Appleby S., 2019, @doi [ ] 10.1093/mnras/stz937 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.486.2827D 486, 2827

    2019 doi

  5. [13]

    arXiv:1807.03819

    Dehghani M., Gouws S., Vinyals O., Uszkoreit J., Kaiser ., 2018, @doi [arXiv e-prints] 10.48550/arXiv.1807.03819 , https://ui.adsabs.harvard.edu/abs/2018arXiv180703819D p. arXiv:1807.03819

    2018 doi arXiv

  6. [14]

    M., Dickey J

    D \'e nes H., McClure-Griffiths N. M., Dickey J. M., Dawson J. R., Murray C. E., 2018, @doi [ ] 10.1093/mnras/sty1384 , http://adsabs.harvard.edu/abs/2018MNRAS.479.1465D 479, 1465

    2018 doi

  7. [15]

    arXiv:1810.04805

    Devlin J., Chang M.-W., Lee K., Toutanova K., 2018, @doi [arXiv e-prints] 10.48550/arXiv.1810.04805 , https://ui.adsabs.harvard.edu/abs/2018arXiv181004805D p. arXiv:1810.04805

    2018 doi arXiv

  8. [16]

    M., Lockman F

    Dickey J. M., Lockman F. J., 1990, @doi [Annual Review of Astronomy and Astrophysics] 10.1146/annurev.aa.28.090190.001243 , 28, 215

    1990

  9. [17]

    M., Salpeter E

    Dickey J. M., Salpeter E. E., Terzian Y., 1978, @doi [ ] 10.1086/190492 , https://ui.adsabs.harvard.edu/abs/1978ApJS...36...77D 36, 77

    1978 doi

  10. [18]

    M., McClure-Griffiths N

    Dickey J. M., McClure-Griffiths N. M., Gaensler B. M., Green A. J., 2003, @doi [ ] 10.1086/346081 , http://adsabs.harvard.edu/abs/2003ApJ...585..801D 585, 801

    2003 doi

  11. [19]

    M., Strasser S., Gaensler B

    Dickey J. M., Strasser S., Gaensler B. M., Haverkorn M., Kavars D., McClure-Griffiths N. M., Stil J., Taylor A. R., 2009, @doi [ ] 10.1088/0004-637X/693/2/1250 , http://adsabs.harvard.edu/abs/2009ApJ...693.1250D 693, 1250

    2009 doi

  12. [20]

    arXiv:2010.11929

    Dosovitskiy A., et al., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2010.11929 , https://ui.adsabs.harvard.edu/abs/2020arXiv201011929D p. arXiv:2010.11929

    2020 doi arXiv

  13. [21]

    arXiv:2102.11090

    Dufter P., Schmitt M., Sch \"u tze H., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2102.11090 , https://ui.adsabs.harvard.edu/abs/2021arXiv210211090D p. arXiv:2102.11090

    2021 doi arXiv

  14. [22]

    Dullemond C. P., Juhasz A., Pohl A., Sereshti F., Shetty R., Peters T., Commercon B., Flock M., 2012, RADMC-3D: A multi-purpose radiative transfer tool , Astrophysics Source Code Library, record ascl:1202.015

    2012

  15. [23]

    B., 1965, @doi [ ] 10.1086/148317 , https://ui.adsabs.harvard.edu/abs/1965ApJ...142..531F 142, 531

    Field G. B., 1965, @doi [ ] 10.1086/148317 , https://ui.adsabs.harvard.edu/abs/1965ApJ...142..531F 142, 531

    1965 doi

  16. [24]

    B., Goldsmith D

    Field G. B., Goldsmith D. W., Habing H. J., 1969, @doi [ ] 10.1086/180324 , http://adsabs.harvard.edu/abs/1969ApJ...155L.149F 155, L149

    1969 doi

  17. [25]

    Fukushima K., 1980, Biological Cybernetics, 36, 193

    1980

  18. [26]

    J., Scalo J., 2001, @doi [ ] 10.1086/322873 , http://adsabs.harvard.edu/abs/2001ApJ...557L.121G 557, L121

    Gazol A., V \'a zquez-Semadeni E., S \'a nchez-Salcedo F. J., Scalo J., 2001, @doi [ ] 10.1086/322873 , http://adsabs.harvard.edu/abs/2001ApJ...557L.121G 557, L121

    2001 doi

  19. [27]

    N., 2017, @doi [arXiv e-prints] 10.48550/arXiv.1705.03122 , https://ui.adsabs.harvard.edu/abs/2017arXiv170503122G p

    Gehring J., Auli M., Grangier D., Yarats D., Dauphin Y. N., 2017, @doi [arXiv e-prints] 10.48550/arXiv.1705.03122 , https://ui.adsabs.harvard.edu/abs/2017arXiv170503122G p. arXiv:1705.03122

    2017 doi arXiv

  20. [28]

    Genel S., et al., 2014, @doi [ ] 10.1093/mnras/stu1654 , https://ui.adsabs.harvard.edu/abs/2014MNRAS.445..175G 445, 175

    2014 doi

  21. [29]

    C., Kim C.-G., 2018, @doi [ ] 10.3847/1538-4357/aab9af , https://ui.adsabs.harvard.edu/abs/2018ApJ...858...16G 858, 16

    Gong M., Ostriker E. C., Kim C.-G., 2018, @doi [ ] 10.3847/1538-4357/aab9af , https://ui.adsabs.harvard.edu/abs/2018ApJ...858...16G 858, 16

    2018 doi

  22. [30]

    C., Kim C.-G., Kim J.-G., 2020, @doi [ ] 10.3847/1538-4357/abbdab , https://ui.adsabs.harvard.edu/abs/2020ApJ...903..142G 903, 142

    Gong M., Ostriker E. C., Kim C.-G., Kim J.-G., 2020, @doi [ ] 10.3847/1538-4357/abbdab , https://ui.adsabs.harvard.edu/abs/2020ApJ...903..142G 903, 142

    2020 doi

  23. [31]

    MIT Press

    Goodfellow I., Bengio Y., Courville A., 2016, Deep Learning. MIT Press

    2016

  24. [32]

    arXiv:2005.08100

    Gulati A., et al., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2005.08100 , https://ui.adsabs.harvard.edu/abs/2020arXiv200508100G p. arXiv:2005.08100

    2020 doi arXiv

  25. [33]

    HI4PI Collaboration et al., 2016, @doi [ ] 10.1051/0004-6361/201629178 , http://adsabs.harvard.edu/abs/2016A

    2016 doi

  26. [34]

    arXiv:2005.03191

    Han W., et al., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2005.03191 , https://ui.adsabs.harvard.edu/abs/2020arXiv200503191H p. arXiv:2005.03191

    2020 doi arXiv

  27. [35]

    Haud U., Kalberla P. M. W., 2007, @doi [ ] 10.1051/0004-6361:20065796 , https://ui.adsabs.harvard.edu/abs/2007A&A...466..555H 466, 555

    2007 doi

  28. [36]

    H., 2003a, @doi [ ] 10.1086/367785 , http://adsabs.harvard.edu/abs/2003ApJS..145..329H 145, 329

    Heiles C., Troland T. H., 2003a, @doi [ ] 10.1086/367785 , http://adsabs.harvard.edu/abs/2003ApJS..145..329H 145, 329

    doi

  29. [37]

    H., 2003b, @doi [ ] 10.1086/367828 , http://adsabs.harvard.edu/abs/2003ApJ...586.1067H 586, 1067

    Heiles C., Troland T. H., 2003b, @doi [ ] 10.1086/367828 , http://adsabs.harvard.edu/abs/2003ApJ...586.1067H 586, 1067

    doi

  30. [38]

    Hennebelle P., P \'e rault M., 2000, , https://ui.adsabs.harvard.edu/abs/2000A&A...359.1124H 359, 1124

    2000

  31. [39]

    S., Murray C

    Hensley B. S., Murray C. E., Dodici M., 2022, @doi [ ] 10.3847/1538-4357/ac5cbd , https://ui.adsabs.harvard.edu/abs/2022ApJ...929...23H 929, 23

    2022 doi

  32. [40]

    S., Mac Low M.-M., Gatto A., Ib \'a \ n ez-Mej \' a J

    Hill A. S., Mac Low M.-M., Gatto A., Ib \'a \ n ez-Mej \' a J. C., 2018, @doi [ ] 10.3847/1538-4357/aacce2 , http://adsabs.harvard.edu/abs/2018ApJ...862...55H 862, 55

    2018 doi

  33. [41]

    D., Krumholz M

    Hu Z., Wibking B. D., Krumholz M. R., 2023, @doi [ ] 10.1093/mnras/stad931 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.521.5604H 521, 5604

    2023 doi

  34. [42]

    D., 2007, @doi [Computing in Science & Engineering] 10.1109/MCSE.2007.55 , 9, 90

    Hunter J. D., 2007, @doi [Computing in Science & Engineering] 10.1109/MCSE.2007.55 , 9, 90

    2007 doi

  35. [43]

    Kalberla P. M. W., Haud U., 2018, @doi [ ] 10.1051/0004-6361/201833146 , http://adsabs.harvard.edu/abs/2018A

    2018 doi

  36. [44]

    arXiv:2006.15595

    Ke G., He D., Liu T.-Y., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2006.15595 , https://ui.adsabs.harvard.edu/abs/2020arXiv200615595K p. arXiv:2006.15595

    2020 doi arXiv

  37. [45]

    Kerp J., Winkel B., Ben Bekhti N., Fl \"o er L., Kalberla P. M. W., 2011, @doi [Astronomische Nachrichten] 10.1002/asna.201011548 , https://ui.adsabs.harvard.edu/abs/2011AN....332..637K 332, 637

    2011 doi

  38. [46]

    arXiv:2101.01169

    Khan S., Naseer M., Hayat M., Waqas Zamir S., Shahbaz Khan F., Shah M., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2101.01169 , https://ui.adsabs.harvard.edu/abs/2021arXiv210101169K p. arXiv:2101.01169

    2021 doi arXiv

  39. [47]

    C., 2017, @doi [ ] 10.3847/1538-4357/aa8599 , https://ui.adsabs.harvard.edu/abs/2017ApJ...846..133K 846, 133

    Kim C.-G., Ostriker E. C., 2017, @doi [ ] 10.3847/1538-4357/aa8599 , https://ui.adsabs.harvard.edu/abs/2017ApJ...846..133K 846, 133

    2017 doi

  40. [48]

    C., Kim W.-T., 2013, @doi [ ] 10.1088/0004-637X/776/1/1 , http://adsabs.harvard.edu/abs/2013ApJ...776....1K 776, 1

    Kim C.-G., Ostriker E. C., Kim W.-T., 2013, @doi [ ] 10.1088/0004-637X/776/1/1 , http://adsabs.harvard.edu/abs/2013ApJ...776....1K 776, 1

    2013 doi

  41. [49]

    C., 2020, @doi [ ] 10.3847/1538-4357/ab9b87 , https://ui.adsabs.harvard.edu/abs/2020ApJ...898...35K 898, 35

    Kim W.-T., Kim C.-G., Ostriker E. C., 2020, @doi [ ] 10.3847/1538-4357/ab9b87 , https://ui.adsabs.harvard.edu/abs/2020ApJ...898...35K 898, 35

    2020 doi

  42. [50]

    C., 2023, @doi [ ] 10.3847/1538-4357/acbd3a , https://ui.adsabs.harvard.edu/abs/2023ApJ...946....3K 946, 3

    Kim C.-G., Kim J.-G., Gong M., Ostriker E. C., 2023, @doi [ ] 10.3847/1538-4357/acbd3a , https://ui.adsabs.harvard.edu/abs/2023ApJ...946....3K 946, 3

    2023 doi

  43. [51]

    arXiv:2001.04451

    Kitaev N., Kaiser ., Levskaya A., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2001.04451 , https://ui.adsabs.harvard.edu/abs/2020arXiv200104451K p. arXiv:2001.04451

    2020 doi arXiv

  44. [52]

    arXiv:1910.10261

    Kriman S., et al., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1910.10261 , https://ui.adsabs.harvard.edu/abs/2019arXiv191010261K p. arXiv:1910.10261

    2019 doi arXiv

  45. [53]

    R., McKee C

    Krumholz M. R., McKee C. F., Tumlinson J., 2009, @doi [ ] 10.1088/0004-637X/693/1/216 , http://adsabs.harvard.edu/abs/2009ApJ...693..216K 693, 216

    2009 doi

  46. [54]

    R., Heiles C., 1988, Neutral hydrogen and the diffuse interstellar medium

    Kulkarni S. R., Heiles C., 1988, Neutral hydrogen and the diffuse interstellar medium . Springer-Verlag, pp 95--153

    1988

  47. [55]

    arXiv:1909.11942

    Lan Z., Chen M., Goodman S., Gimpel K., Sharma P., Soricut R., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1909.11942 , https://ui.adsabs.harvard.edu/abs/2019arXiv190911942L p. arXiv:1909.11942

    2019 doi arXiv

  48. [56]

    Lecun Y., Bottou L., Bengio Y., Haffner P., 1998, @doi [Proceedings of the IEEE] 10.1109/5.726791 , 86, 2278

    1998 doi

  49. [57]

    E., Heiles C., Miller J., 2015, @doi [ ] 10.1088/0004-637X/809/1/56 , http://adsabs.harvard.edu/abs/2015ApJ...809...56L 809, 56

    Lee M.-Y., Stanimirovi \'c S., Murray C. E., Heiles C., Miller J., 2015, @doi [ ] 10.1088/0004-637X/809/1/56 , http://adsabs.harvard.edu/abs/2015ApJ...809...56L 809, 56

    2015 doi

  50. [58]

    E., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2312.03846 , https://ui.adsabs.harvard.edu/abs/2023arXiv231203846L p

    Lei M., Clark S. E., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2312.03846 , https://ui.adsabs.harvard.edu/abs/2023arXiv231203846L p. arXiv:2312.03846

    2023 doi arXiv

  51. [59]

    M., Nguyen H., Teja Gadde R., 2019a, @doi [arXiv e-prints] 10.48550/arXiv.1904.03288 , https://ui.adsabs.harvard.edu/abs/2019arXiv190403288L p

    Li J., Lavrukhin V., Ginsburg B., Leary R., Kuchaiev O., Cohen J. M., Nguyen H., Teja Gadde R., 2019a, @doi [arXiv e-prints] 10.48550/arXiv.1904.03288 , https://ui.adsabs.harvard.edu/abs/2019arXiv190403288L p. arXiv:1904.03288

    1904 doi arXiv

  52. [60]

    Li H., Wang A. Y. C., Liu Y., Tang D., Lei Z., Li W., 2019b, @doi [arXiv e-prints] 10.48550/arXiv.1910.13634 , https://ui.adsabs.harvard.edu/abs/2019arXiv191013634L p. arXiv:1910.13634

    1910 doi arXiv

  53. [61]

    Liszt H., 2001, @doi [ ] 10.1051/0004-6361:20010395 , http://adsabs.harvard.edu/abs/2001A

    2001 doi

  54. [62]

    arXiv:1907.11692

    Liu Y., et al., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1907.11692 , https://ui.adsabs.harvard.edu/abs/2019arXiv190711692L p. arXiv:1907.11692

    2019 doi arXiv

  55. [63]

    arXiv:1906.02762

    Lu Y., Li Z., He D., Sun Z., Dong B., Qin T., Wang L., Liu T.-Y., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1906.02762 , https://ui.adsabs.harvard.edu/abs/2019arXiv190602762L p. arXiv:1906.02762

    2019 doi arXiv

  56. [64]

    R., Salom \'e Q., 2019, @doi [ ] 10.1051/0004-6361/201935335 , https://ui.adsabs.harvard.edu/abs/2019A&A...626A.101M 626, A101

    Marchal A., Miville-Desch \^e nes M.-A., Orieux F., Gac N., Soussen C., Lesot M.-J., d'Allonnes A. R., Salom \'e Q., 2019, @doi [ ] 10.1051/0004-6361/201935335 , https://ui.adsabs.harvard.edu/abs/2019A&A...626A.101M 626, A101

    2019 doi

  57. [65]

    G., Miville-Desch \^e nes M.-A., McClure-Griffiths N

    Marchal A., Martin P. G., Miville-Desch \^e nes M.-A., McClure-Griffiths N. M., Lynn C., Bracco A., Vujeva L., 2024, @doi [ ] 10.3847/1538-4357/ad0f21 , https://ui.adsabs.harvard.edu/abs/2024ApJ...961..161M 961, 161

    2024 doi

  58. [66]

    G., Blagrave K

    Martin P. G., Blagrave K. P. M., Lockman F. J., Pinheiro Gon c alves D., Boothroyd A. I., Joncas G., Miville-Desch \^e nes M. A., Stephan G., 2015, @doi [ ] 10.1088/0004-637X/809/2/153 , https://ui.adsabs.harvard.edu/abs/2015ApJ...809..153M 809, 153

    2015 doi

  59. [67]

    M., et al., 2009, @doi [ ] 10.1088/0067-0049/181/2/398 , https://ui.adsabs.harvard.edu/abs/2009ApJS..181..398M 181, 398

    McClure-Griffiths N. M., et al., 2009, @doi [ ] 10.1088/0067-0049/181/2/398 , https://ui.adsabs.harvard.edu/abs/2009ApJS..181..398M 181, 398

    2009 doi

  60. [68]

    M., et al., 2015, in Advancing Astrophysics with the Square Kilometre Array (AASKA14)

    McClure-Griffiths N. M., et al., 2015, in Advancing Astrophysics with the Square Kilometre Array (AASKA14). p. 130 ( @eprint arXiv 1501.01130 ), @doi 10.22323/1.215.0130

    2015 arXiv

  61. [69]

    M., Stanimirovi \'c S., Rybarczyk D

    McClure-Griffiths N. M., Stanimirovi \'c S., Rybarczyk D. R., 2023, @doi [ ] 10.1146/annurev-astro-052920-104851 , https://ui.adsabs.harvard.edu/abs/2023ARA&A..61...19M 61, 19

    2023 doi

  62. [70]

    F., Ostriker J

    McKee C. F., Ostriker J. P., 1977, @doi [ ] 10.1086/155667 , http://adsabs.harvard.edu/abs/1977ApJ...218..148M 218, 148

    1977 doi

  63. [71]

    pp 51--56

    McKinney W., et al., 2010, in Proceedings of the 9th Python in Science Conference. pp 51--56

    2010

  64. [72]

    Mebold U., Winnberg A., Kalberla P. M. W., Goss W. M., 1982, , http://adsabs.harvard.edu/abs/1982A

    1982

  65. [73]

    S., Srinivasan G., 2004, @doi [Journal of Astrophysics and Astronomy] 10.1007/BF02702370 , https://ui.adsabs.harvard.edu/abs/2004JApA...25..143M 25, 143

    Mohan R., Dwarakanath K. S., Srinivasan G., 2004, @doi [Journal of Astrophysics and Astronomy] 10.1007/BF02702370 , https://ui.adsabs.harvard.edu/abs/2004JApA...25..143M 25, 143

    2004 doi

  66. [74]

    E., et al., 2015, @doi [ ] 10.1088/0004-637X/804/2/89 , http://adsabs.harvard.edu/abs/2015ApJ...804...89M 804, 89

    Murray C. E., et al., 2015, @doi [ ] 10.1088/0004-637X/804/2/89 , http://adsabs.harvard.edu/abs/2015ApJ...804...89M 804, 89

    2015 doi

  67. [75]

    E., Stanimirovi \'c S., Goss W

    Murray C. E., Stanimirovi \'c S., Goss W. M., Heiles C., Dickey J. M., Babler B., Kim C.-G., 2018a, @doi [ ] 10.3847/1538-4365/aad81a , http://adsabs.harvard.edu/abs/2018ApJS..238...14M 238, 14

    doi

  68. [76]

    E., Peek J

    Murray C. E., Peek J. E. G., Lee M.-Y., Stanimirovi \'c S., 2018b, @doi [ ] 10.3847/1538-4357/aaccfe , http://adsabs.harvard.edu/abs/2018ApJ...862..131M 862, 131

    doi

  69. [77]

    E., Peek J

    Murray C. E., Peek J. E. G., Kim C.-G., 2020, @doi [ ] 10.3847/1538-4357/aba19b , https://ui.adsabs.harvard.edu/abs/2020ApJ...899...15M 899, 15

    2020 doi

  70. [78]

    E., Stanimirovi \'c S., Heiles C., Dickey J

    Murray C. E., Stanimirovi \'c S., Heiles C., Dickey J. M., McClure-Griffiths N. M., Lee M. Y., M. Goss W., Killerby-Smith N., 2021, @doi [ ] 10.3847/1538-4365/ac0f0b , https://ui.adsabs.harvard.edu/abs/2021ApJS..256...37M 256, 37

    2021 doi

  71. [79]

    Nguyen H., et al., 2018, @doi [ ] 10.3847/1538-4357/aac82b , http://adsabs.harvard.edu/abs/2018ApJ...862...49N 862, 49

    2018 doi

  72. [80]

    R., Lee M.-Y., Murray C

    Nguyen H., Dawson J. R., Lee M.-Y., Murray C. E., Stanimirovi \'c S., Heiles C., Miville-Desch \^e nes M. A., Petzler A., 2019, @doi [ ] 10.3847/1538-4357/ab2b9f , https://ui.adsabs.harvard.edu/abs/2019ApJ...880..141N 880, 141

    2019 doi

  73. [81]

    Nguyen H., et al., 2024, @doi [ ] 10.1093/mnras/stae2274 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.534.3478N 534, 3478

    2024 doi

  74. [82]

    Pan J., Ting Y.-S., Yu J., 2022, in Machine Learning for Astrophysics. p. 10 ( @eprint arXiv 2207.02787 ), @doi 10.48550/arXiv.2207.02787

    2022 doi arXiv

  75. [83]

    arXiv:1912.01703

    Paszke A., et al., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1912.01703 , https://ui.adsabs.harvard.edu/abs/2019arXiv191201703P p. arXiv:1912.01703

    2019 doi arXiv

  76. [84]

    Peek J. E. G., et al., 2011, @doi [ ] 10.1088/0067-0049/194/2/20 , http://adsabs.harvard.edu/abs/2011ApJS..194...20P 194, 20

    2011 doi

  77. [85]

    Peek J. E. G., et al., 2018, @doi [ ] 10.3847/1538-4365/aa91d3 , http://adsabs.harvard.edu/abs/2018ApJS..234....2P 234, 2

    2018 doi

  78. [86]

    Radford A., Wu J., Child R., Luan D., Amodei D., Sutskever I., 2018, https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf

    2018

  79. [87]

    arXiv:1910.10683

    Raffel C., et al., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1910.10683 , https://ui.adsabs.harvard.edu/abs/2019arXiv191010683R p. arXiv:1910.10683

    2019 doi arXiv

  80. [88]

    N., 2013a, @doi [ ] 10.1093/mnras/stt1743 , http://adsabs.harvard.edu/abs/2013MNRAS.436.2352R 436, 2352

    Roy N., Kanekar N., Braun R., Chengalur J. N., 2013a, @doi [ ] 10.1093/mnras/stt1743 , http://adsabs.harvard.edu/abs/2013MNRAS.436.2352R 436, 2352

    doi

  81. [89]

    N., 2013b, @doi [ ] 10.1093/mnras/stt1746 , https://ui.adsabs.harvard.edu/abs/2013MNRAS.436.2366R 436, 2366

    Roy N., Kanekar N., Chengalur J. N., 2013b, @doi [ ] 10.1093/mnras/stt1746 , https://ui.adsabs.harvard.edu/abs/2013MNRAS.436.2366R 436, 2366

    doi

  82. [90]

    arXiv:2306.15703

    R \'o \.z a \'n ski T., Ting Y.-S., Jab o \'n ska M., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2306.15703 , https://ui.adsabs.harvard.edu/abs/2023arXiv230615703R p. arXiv:2306.15703

    2023 doi arXiv

  83. [91]

    N., Mohamed A.-r., Kingsbury B., Ramabhadran B., 2013, in 2013 IEEE International Conference on Acoustics, Speech and Signal Processing

    Sainath T. N., Mohamed A.-r., Kingsbury B., Ramabhadran B., 2013, in 2013 IEEE International Conference on Acoustics, Speech and Signal Processing. pp 8614--8618, @doi 10.1109/ICASSP.2013.6639347

    2013

  84. [92]

    Saury E., Miville-Desch \^e nes M.-A., Hennebelle P., Audit E., Schmidt W., 2014, @doi [ ] 10.1051/0004-6361/201321113 , http://adsabs.harvard.edu/abs/2014A

    2014 doi

  85. [93]

    Schaye J., et al., 2015, @doi [ ] 10.1093/mnras/stu2058 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.446..521S 446, 521

    2015 doi

  86. [94]

    Seta A., Federrath C., 2022, @doi [ ] 10.1093/mnras/stac1400 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.514..957S 514, 957

    2022 doi

  87. [95]

    J., Hubeny I., 2017, @doi [ ] 10.3847/1538-4357/aa7747 , http://adsabs.harvard.edu/abs/2017ApJ...843..149S 843, 149

    Shaw G., Ferland G. J., Hubeny I., 2017, @doi [ ] 10.3847/1538-4357/aa7747 , http://adsabs.harvard.edu/abs/2017ApJ...843..149S 843, 149

    2017 doi

  88. [96]

    Springel V., et al., 2018, @doi [ ] 10.1093/mnras/stx3304 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.475..676S 475, 676

    2018 doi

  89. [97]

    E., Lee M.-Y., Heiles C., Miller J., 2014, @doi [ ] 10.1088/0004-637X/793/2/132 , http://adsabs.harvard.edu/abs/2014ApJ...793..132S 793, 132

    Stanimirovi \'c S., Murray C. E., Lee M.-Y., Heiles C., Miller J., 2014, @doi [ ] 10.1088/0004-637X/793/2/132 , http://adsabs.harvard.edu/abs/2014ApJ...793..132S 793, 132

    2014 doi

  90. [98]

    T., et al., 2007, @doi [ ] 10.1086/522794 , https://ui.adsabs.harvard.edu/abs/2007AJ....134.2252S 134, 2252

    Strasser S. T., et al., 2007, @doi [ ] 10.1086/522794 , https://ui.adsabs.harvard.edu/abs/2007AJ....134.2252S 134, 2252

    2007 doi

  91. [99]

    pp 1--9, @doi 10.1109/CVPR.2015.7298594

    Szegedy C., et al., 2015a, in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp 1--9, @doi 10.1109/CVPR.2015.7298594

    2015

  92. [100]

    arXiv:1512.00567

    Szegedy C., Vanhoucke V., Ioffe S., Shlens J., Wojna Z., 2015b, @doi [arXiv e-prints] 10.48550/arXiv.1512.00567 , https://ui.adsabs.harvard.edu/abs/2015arXiv151200567S p. arXiv:1512.00567

    doi arXiv

  93. [101]

    arXiv:2308.11455

    Uelwer T., Robine J., Sylvius Wagner S., H \"o ftmann M., Upschulte E., Konietzny S., Behrendt M., Harmeling S., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2308.11455 , https://ui.adsabs.harvard.edu/abs/2023arXiv230811455U p. arXiv:2308.11455

    2023 doi arXiv

  94. [102]

    N., Kaiser L., Polosukhin I., 2017, @doi [arXiv e-prints] 10.48550/arXiv.1706.03762 , https://ui.adsabs.harvard.edu/abs/2017arXiv170603762V p

    Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A. N., Kaiser L., Polosukhin I., 2017, @doi [arXiv e-prints] 10.48550/arXiv.1706.03762 , https://ui.adsabs.harvard.edu/abs/2017arXiv170603762V p. arXiv:1706.03762

    2017 doi arXiv

  95. [103]

    R., Wibking B

    Vijayan A., Krumholz M. R., Wibking B. D., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2309.07955 , https://ui.adsabs.harvard.edu/abs/2023arXiv230907955V p. arXiv:2309.07955

    2023 doi arXiv

  96. [104]

    Virtanen P., et al., 2020, @doi [Nature Methods] 10.1038/s41592-019-0686-2 , https://ui.adsabs.harvard.edu/abs/2020NatMe..17..261V 17, 261

    2020 doi

  97. [105]

    Vogelsberger M., et al., 2014, @doi [ ] 10.1093/mnras/stu1536 , https://ui.adsabs.harvard.edu/abs/2014MNRAS.444.1518V 444, 1518

    2014 doi

  98. [106]

    G., Taank M., 2023, @doi [ ] 10.3847/1538-4357/acd340 , https://ui.adsabs.harvard.edu/abs/2023ApJ...951..120V 951, 120

    Vujeva L., Marchal A., Martin P. G., Taank M., 2023, @doi [ ] 10.3847/1538-4357/acd340 , https://ui.adsabs.harvard.edu/abs/2023ApJ...951..120V 951, 120

    2023 doi

  99. [107]

    P., 2001, @doi [ ] 10.1086/321783 , https://ui.adsabs.harvard.edu/abs/2001ApJS..136..463W 136, 463

    Wakker B. P., 2001, @doi [ ] 10.1086/321783 , https://ui.adsabs.harvard.edu/abs/2001ApJS..136..463W 136, 463

    2001 doi

  100. [108]

    Winkel B., Kalberla P. M. W., Kerp J., Fl \"o er L., 2010, @doi [ ] 10.1088/0067-0049/188/2/488 , https://ui.adsabs.harvard.edu/abs/2010ApJS..188..488W 188, 488

    2010 doi

  101. [109]

    Winkel B., Kerp J., Fl \"o er L., Kalberla P. M. W., Ben Bekhti N., Keller R., Lenz D., 2016, @doi [ ] 10.1051/0004-6361/201527007 , https://ui.adsabs.harvard.edu/abs/2016A&A...585A..41W 585, A41

    2016 doi

  102. [110]

    G., Hollenbach D., McKee C

    Wolfire M. G., Hollenbach D., McKee C. F., Tielens A. G. G. M., Bakes E. L. O., 1995, @doi [ ] 10.1086/175510 , http://adsabs.harvard.edu/abs/1995ApJ...443..152W 443, 152

    1995 doi

  103. [111]

    G., McKee C

    Wolfire M. G., McKee C. F., Hollenbach D., Tielens A. G. G. M., 2003, @doi [ ] 10.1086/368016 , http://adsabs.harvard.edu/abs/2003ApJ...587..278W 587, 278

    2003 doi

  104. [112]

    arXiv:2004.11886

    Wu Z., Liu Z., Lin J., Lin Y., Han S., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2004.11886 , https://ui.adsabs.harvard.edu/abs/2020arXiv200411886W p. arXiv:2004.11886

    2020 doi arXiv

  105. [113]

    S., Tu Z., 2019a, @doi [arXiv e-prints] 10.48550/arXiv.1904.03107 , https://ui.adsabs.harvard.edu/abs/2019arXiv190403107Y p

    Yang B., Wang L., Wong D., Chao L. S., Tu Z., 2019a, @doi [arXiv e-prints] 10.48550/arXiv.1904.03107 , https://ui.adsabs.harvard.edu/abs/2019arXiv190403107Y p. arXiv:1904.03107

    1904 doi arXiv

  106. [114]

    V., 2019b, @doi [arXiv e-prints] 10.48550/arXiv.1906.08237 , https://ui.adsabs.harvard.edu/abs/2019arXiv190608237Y p

    Yang Z., Dai Z., Yang Y., Carbonell J., Salakhutdinov R., Le Q. V., 2019b, @doi [arXiv e-prints] 10.48550/arXiv.1906.08237 , https://ui.adsabs.harvard.edu/abs/2019arXiv190608237Y p. arXiv:1906.08237

    1906 doi arXiv

  107. [115]

    W., Dohan D., Luong M.-T., Zhao R., Chen K., Norouzi M., Le Q

    Yu A. W., Dohan D., Luong M.-T., Zhao R., Chen K., Norouzi M., Le Q. V., 2018, @doi [arXiv e-prints] 10.48550/arXiv.1804.09541 , https://ui.adsabs.harvard.edu/abs/2018arXiv180409541Y p. arXiv:1804.09541

    2018 doi arXiv

  108. [116]

    arXiv:2002.02562

    Zhang Q., Lu H., Sak H., Tripathi A., McDermott E., Koo S., Kumar S., 2020, @doi [arXiv e-prints] 10.48550/arXiv.2002.02562 , https://ui.adsabs.harvard.edu/abs/2020arXiv200202562Z p. arXiv:2002.02562

    2020 doi arXiv

  109. [117]

    C., Varoquaux G., 2011, @doi [Computing in Science and Engineering] 10.1109/MCSE.2011.37 , https://ui.adsabs.harvard.edu/abs/2011CSE....13b..22V 13, 22

    van der Walt S., Colbert S. C., Varoquaux G., 2011, @doi [Computing in Science and Engineering] 10.1109/MCSE.2011.37 , https://ui.adsabs.harvard.edu/abs/2011CSE....13b..22V 13, 22

    2011 doi

Pith tools

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