pith. machine review for the scientific record. sign in

arxiv: 2604.19919 · v1 · submitted 2026-04-21 · 🌌 astro-ph.SR · astro-ph.GA· astro-ph.IM

Recognition: unknown

Neural Simulation-based Inference with Hierarchical Priors for Detached Eclipsing Binaries

Jacqueline Blaum Hough , Joshua S. Bloom
Authors on Pith no claims yet

Pith reviewed 2026-05-10 01:15 UTC · model grok-4.3

classification 🌌 astro-ph.SR astro-ph.GAastro-ph.IM
keywords detached eclipsing binariesamortized inferenceneural posterior estimationsimulation-based inferencelight curve analysisMIST isochronesstellar parametershierarchical priors
0
0 comments X

The pith

A neural network approximates the full posterior over 16 parameters of detached eclipsing binaries from light curves, SEDs and parallaxes alone.

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

The paper shows that an amortized neural posterior estimator, trained once on large sets of simulated detached eclipsing binary systems, can recover accurate joint posteriors for stellar masses, radii, temperatures, orbital elements and population parameters. The training data are generated under a hierarchical prior that enforces consistency with MIST isochrones while including realistic survey cadence and noise. On nearly 5000 held-out simulations the recovered posteriors are well calibrated according to simulation-based calibration checks and coverage tests. Quantities fixed by eclipse geometry are tightly constrained while age and metallicity remain broader, as expected from photometric degeneracies. Because inference on a new system is essentially instantaneous after the one-time training cost, the approach removes the computational barrier that has limited homogeneous parameter catalogs to small numbers of systems.

Core claim

The authors introduce multimodal amortized neural posterior estimation that combines survey-realistic light curves, broadband SEDs and Gaia parallaxes inside a physically motivated hierarchical prior. A conditional normalizing flow, fed by modality-specific encoders, directly approximates the 16-dimensional posterior. The generative model enforces MIST isochrone consistency and geometric eclipse constraints while injecting empirically derived cadence patterns and flux-dependent noise. Across thousands of held-out simulations the method recovers parameters accurately and produces statistically calibrated uncertainties, with geometry- and flux-linked quantities tightly constrained and age-met

What carries the argument

Conditional normalizing flow that approximates the joint posterior from encoded multimodal observations under a hierarchical prior enforcing MIST isochrone consistency and geometric eclipse constraints.

If this is right

  • Parameters tied to eclipse geometry and overall flux scale are recovered with high precision while age and metallicity posteriors remain appropriately broad.
  • The cost of generating the entire training set is comparable to a single traditional MCMC analysis of one system.
  • Posterior evaluation for any new system is effectively instantaneous after training.
  • The framework directly supports population-level statistical studies of large photometrically selected DEB catalogs without requiring radial-velocity follow-up for every target.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same amortized approach could be retrained for other classes of variable stars or transiting planets once suitable generative models exist.
  • Catalogs produced by this method would enable direct comparison of observed binary populations against binary evolution simulations across wide ranges of age and metallicity.
  • When sparse radial-velocity measurements become available for a subset of systems they could be added as an additional conditioning modality to shrink the remaining degeneracies.

Load-bearing premise

The simulated light curves, SEDs and noise properties used for training are statistically close enough to real observations that the network generalizes without large domain shift.

What would settle it

Apply the trained network to a sample of real detached eclipsing binaries that already have independent MCMC or RV-based parameter estimates; systematic offsets in the recovered means or failure of the reported uncertainties to cover the true values at the expected rate would falsify the claim.

Figures

Figures reproduced from arXiv: 2604.19919 by Jacqueline Blaum Hough, Joshua S. Bloom.

Figure 1
Figure 1. Figure 1: Overview of the hierarchical generative model used to construct the simulation-based training dataset. A population model defines priors over the 16-dimensional parameter vector θ, which specifies a physically consistent detached eclipsing binary configuration. These parameters are passed to the PHOEBE forward model to generate noiseless light curves and spectral energy distributions (SEDs). Survey-specifi… view at source ↗
Figure 2
Figure 2. Figure 2: Hierarchical construction of the 16-dimensional parameter vector θ. Distance and reddening are drawn from a joint empirical prior based on Gaia cross-matched eclipsing binaries and the Bayestar three-dimensional dust map. Core priors define fundamental stellar and orbital quantities (e.g., total mass, mass ratio, age, metallicity, orbital elements), which are transformed into physically consistent stellar … view at source ↗
Figure 3
Figure 3. Figure 3: One- and two-dimensional distributions of the effective priors on stellar parameters. A small number of samples lie outside the plotted ranges and are omitted for clarity. 2.2.2. Spectral Energy Distributions We generate synthetic broadband SEDs by computing band-integrated fluxes in a fixed set of 23 canonical passbands spanning GALEX (UV) through WISE (mid-IR). The full filter list can be found in [PITH… view at source ↗
Figure 4
Figure 4. Figure 4: One- and two-dimensional marginal distributions of the effective priors on binary parameters (left) and orbital parameters (right), shown as smoothed density contours. The contours emphasize regions of highest prior probability and do not reflect the full extent of the prior support. A small number of samples lie outside the plotted ranges and are omitted for clarity. Distance Scaling. —The PHOEBE bundle i… view at source ↗
Figure 5
Figure 5. Figure 5: Schematic overview of the multimodal ANPE. Simulated observations x consist of survey-realistic light curves (ASAS-SN and ZTF), reddened broadband SEDs, and auxiliary metadata (e.g., orbital period and parallax). Each modality is processed by a dedicated featurizer network (ResNet encoders for light curves and SEDs; GRU for metadata), producing a shared latent embedding h. This embedding conditions a norma… view at source ↗
Figure 6
Figure 6. Figure 6: Recovery of simulated parameters over 4999 trials. For each parameter, green points show the posterior median versus the true value. The solid line indicates the one-to-one relation. Shaded regions show the binned 68% (16th–84th percentile; orange) and 90% (5th–95th percentile; blue) central credible intervals as a function of the true parameter value. ω becomes effectively unconstrained. The observed scat… view at source ↗
Figure 7
Figure 7. Figure 7: Simulation-based calibration (SBC) rank histograms for all inferred parameters, computed over 4999 simulated systems. For each trial, the true parameter value is ranked among posterior samples drawn from the amortized neural posterior. Under perfect calibration, the rank distribution is uniform (dashed horizontal line). All parameters show approximately flat rank histograms with no systematic U-shape or ed… view at source ↗
Figure 8
Figure 8. Figure 8: Empirical marginal coverage across 4999 simulated systems. For each parameter, we compute central credible intervals at the 50%, 68%, 90%, and 95% levels and measure the fraction of trials in which the true value lies within the interval. Error bars indicate the binomial standard error on the estimated covered fraction, SE = p pˆ(1 − pˆ)/N with N = 4999. Dashed horizontal lines indicate the nominal coverag… view at source ↗
Figure 9
Figure 9. Figure 9: Posterior predictive checks for a system from the held-out test set (Test System 1). Normalized light curves are shown on the left and are offset for visibility of each passband, with corresponding residuals (data minus model) shown in the lower panel. The broadband SED is shown on the right, with fractional residuals in log space (∆ log10 Fν) shown below. In all panels, the solid black curve denotes the m… view at source ↗
Figure 10
Figure 10. Figure 10: Posterior corner plots for stellar and orbital parameters of Test System 1. The contours correspond to the 68% and 95% highest posterior density regions. Titles on the diagonal panels list the posterior median and central 68% credible interval. The remaining posteriors are shown in [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Posterior corner plots for the remaining parameters of Test System 1 not shown in [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Same as [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Same as [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Same as [PITH_FULL_IMAGE:figures/full_fig_p023_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Same as Figures 9 and 12 for Test System 3. The ablation results also clarify how different modalities combine to constrain stellar properties. The stellar radii degrade significantly when any modality is removed, indicating that they require the combined constraints of eclipse geometry, stellar properties, and distance-dependent flux scaling. In contrast, the mass ratio remains comparatively robust in co… view at source ↗
Figure 16
Figure 16. Figure 16: Same as Figures 10 and 13 for Test System 3. depending on the system and convergence behavior. By contrast, ANPE posterior inference requires ∼0.45 ms on a single NVIDIA A100 GPU. ANPE can therefore be applied at scale to efficiently infer posterior distributions for [PITH_FULL_IMAGE:figures/full_fig_p025_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Same as Figures 11 and 14 for Test System 3. large samples of DEBs, enabling population-level analyses that would be computationally prohibitive with traditional MCMC methods. This distinction is particularly significant in the era of large time-domain surveys. While traditional MCMC ap￾proaches remain well suited for detailed modeling of individual systems, amortized inference offers a scalable alternati… view at source ↗
Figure 18
Figure 18. Figure 18: Results of the ablation study showing the relative posterior uncertainty for each parameter and input configuration. Each cell reports the ratio of the median 68% credible interval width for a given ablation configuration relative to the full model (LCs+SEDs+Meta), indicated both numerically and through the logarithmic color scale. Values greater than unity indicate degraded constraints, while values belo… view at source ↗
Figure 19
Figure 19. Figure 19: Results of the ablation study showing the posterior accuracy for each parameter and input configuration. Each cell reports the median normalized bias, defined as (q50 −θtrue)/σ, for a given ablation configuration, indicated both numerically and through the color scale. Values near zero indicate unbiased estimates, while positive (negative) values correspond to systematic overestimation (underestimation) o… view at source ↗
Figure 20
Figure 20. Figure 20: Examples of discretization artifacts in reported photometric uncertainties. The panels show photometric uncer￾tainty as a function of magnitude for two surveys with particularly strong effects: the 2MASS H band (left) and the WISE W1 band (right). In both cases, the uncertainties appear in narrow horizontal bands, producing a “barcode-like” pattern. of photometric uncertainty as a function of magnitude, t… view at source ↗
read the original abstract

Detached eclipsing binaries (DEBs) enable direct inference of stellar and orbital properties across diverse stellar populations. However, inference typically requires computationally intensive forward modeling and radial velocity (RV) measurements, limiting homogeneous analyses to relatively small samples. The growing number of photometrically identified DEBs from modern time-domain surveys motivates scalable methods for extracting physical parameters without RVs. We present multimodal amortized neural posterior estimation for DEB inference that combines survey-realistic light curves, broadband SEDs, and Gaia parallaxes within a physically motivated hierarchical prior framework. The generative model enforces broad stellar evolution consistency through MIST isochrones and geometric eclipse prior constraints while incorporating empirically derived survey cadence patterns and flux-dependent noise models to produce realistic training data. A conditional normalizing flow, informed by modality-specific encoders, approximates the full 16-dimensional posterior distribution. Across nearly 5000 held-out simulations, the amortized posterior recovers parameters accurately and yields statistically calibrated uncertainties, verified through simulation-based calibration and empirical coverage tests. Parameters tied directly to eclipse geometry and flux scale are tightly constrained, while quantities intrinsically degenerate in broadband photometry (e.g., age and metallicity) exhibit broader posteriors consistent with expectations. Generating the training set requires computational effort similar to a traditional MCMC analysis of only a single system, and posterior inference for new systems is effectively instantaneous. This framework enables scalable, statistically calibrated inference for large DEB samples, providing a pathway toward population-level analysis in the era of large time-domain surveys.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper presents a multimodal amortized neural posterior estimation framework using conditional normalizing flows to infer 16-dimensional posteriors for detached eclipsing binary (DEB) parameters from survey light curves, broadband SEDs, and Gaia parallaxes. It employs a generative model with MIST isochrone-based hierarchical priors, geometric eclipse constraints, and empirically derived cadence/noise models to produce realistic training simulations, demonstrating accurate parameter recovery and statistically calibrated uncertainties on nearly 5000 held-out simulations via simulation-based calibration (SBC) and empirical coverage tests.

Significance. If the results hold, this provides a scalable, computationally efficient alternative to traditional MCMC for analyzing large photometrically identified DEB samples without radial velocities, enabling population-level studies in the era of time-domain surveys. Credit is due for the use of SBC and coverage diagnostics to verify calibration within the simulated domain, and for the amortized approach that reduces per-system inference to near-instantaneous after upfront training cost comparable to a single MCMC run.

major comments (2)
  1. [Abstract and validation results] Abstract and validation results: the central claim that the framework enables scalable, calibrated inference for real survey DEBs rests on the untested assumption that the MIST-based generative model (with flux-dependent noise and cadence) produces training data sufficiently representative of observations; no quantitative comparison of simulated vs. observed distributions (e.g., eclipse depth/period histograms, SED shapes, or parallax-flux relations) or application to any real DEB system is provided, leaving potential domain shift from unmodeled effects (spots, third light, non-MIST evolution) unaddressed.
  2. [Generative model and prior section] Generative model and prior section: while the hierarchical prior enforces MIST isochrone consistency and geometric constraints, the manuscript does not report metrics (e.g., Kolmogorov-Smirnov tests or posterior predictive checks) quantifying how well the simulated population matches real DEB catalogs, which is load-bearing for transferring the reported calibration to actual data.
minor comments (2)
  1. [Methods] Notation for the 16-dimensional parameter vector and modality-specific encoders could be introduced earlier with a clear table to improve readability for readers unfamiliar with the exact parameterization.
  2. [Results] The abstract states 'nearly 5000 held-out simulations' but the precise number, selection criteria, and any stratification by parameter ranges should be stated explicitly in the results section for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their positive evaluation of the work's significance and for the constructive comments, which correctly identify the need for stronger support when transferring simulation-based results to real observations. We respond point-by-point below and will incorporate revisions to add quantitative comparisons, statistical metrics, and an explicit limitations discussion while preserving the paper's focus on methodological validation within the simulated domain.

read point-by-point responses

  1. Referee: [Abstract and validation results] Abstract and validation results: the central claim that the framework enables scalable, calibrated inference for real survey DEBs rests on the untested assumption that the MIST-based generative model (with flux-dependent noise and cadence) produces training data sufficiently representative of observations; no quantitative comparison of simulated vs. observed distributions (e.g., eclipse depth/period histograms, SED shapes, or parallax-flux relations) or application to any real DEB system is provided, leaving potential domain shift from unmodeled effects (spots, third light, non-MIST evolution) unaddressed.

    Authors: We agree that the absence of direct comparisons to observed DEB distributions and real-system applications leaves the transferability of the reported calibration untested, which is a substantive limitation for claims about survey applicability. The manuscript's scope is the introduction and simulation-based validation of the amortized NPE framework, with SBC and coverage tests providing calibration diagnostics under the assumed generative model (standard for SBI). In revision we will: add a limitations subsection explicitly discussing unmodeled effects (spots, third light, non-MIST evolution) and their potential impact on domain shift; include qualitative and quantitative comparisons (histograms and KS tests) of simulated vs. observed distributions for periods, eclipse depths, SED shapes, and parallax-flux relations using public catalogs (e.g., Kepler EB catalog); and revise the abstract and conclusions to state that calibration holds within the simulated domain, with real-data applications requiring additional systematics handling and planned as follow-up. Full re-analysis of real systems is beyond the current methodological focus but can be noted as future work. revision: partial

  2. Referee: [Generative model and prior section] Generative model and prior section: while the hierarchical prior enforces MIST isochrone consistency and geometric constraints, the manuscript does not report metrics (e.g., Kolmogorov-Smirnov tests or posterior predictive checks) quantifying how well the simulated population matches real DEB catalogs, which is load-bearing for transferring the reported calibration to actual data.

    Authors: We concur that explicit quantitative metrics are needed to assess generative-model fidelity and support transfer of calibration results. The hierarchical prior and empirically derived noise/cadence components were chosen for broad consistency with stellar evolution and survey data, but without reported statistics the match remains qualitative. In the revised manuscript we will add a new subsection reporting KS tests (with p-values) and posterior predictive checks on key observables including orbital period, eclipse depth, effective temperature, surface gravity, and color-magnitude distributions, benchmarked against reference DEB catalogs. These additions will directly quantify the simulated population's fidelity and clarify the conditions under which the reported calibration can be expected to hold for actual survey data. revision: yes

Circularity Check

0 steps flagged

No significant circularity; results are empirical validation on held-out simulations

full rationale

The paper trains a conditional normalizing flow on data generated from an explicit forward model (MIST isochrones, geometric eclipse constraints, empirical noise) and reports recovery accuracy plus calibration metrics exclusively on a separate held-out simulation set. No step equates a claimed prediction to a fitted quantity by construction, no load-bearing self-citation chain is invoked to justify the central result, and the generative model is stated as an input rather than derived from the inference outputs. The evaluation therefore remains independent of the target data and does not reduce to a tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on established stellar evolution models and simulation techniques; specific implementation details such as exact network hyperparameters or noise model parameters are not provided in the abstract.

axioms (1)
  • domain assumption MIST isochrones provide a sufficiently accurate representation of stellar evolution for generating plausible DEB systems
    Invoked to enforce broad stellar evolution consistency in the generative model.

pith-pipeline@v0.9.0 · 5570 in / 1269 out tokens · 38615 ms · 2026-05-10T01:15:30.827215+00:00 · methodology

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

24 extracted references · 24 canonical work pages

  1. [1]

    1991, A&A Rv, 3, 91, doi: 10.1007/BF00873538

    Andersen, J. 1991, A&A Rv, 3, 91, doi: 10.1007/BF00873538

    work page doi:10.1007/bf00873538 1991

  2. [2]

    C., Kulkarni, S

    Bellm, E. C., Kulkarni, S. R., Graham, M. J., et al. 2019, PASP, 131, 018002, doi: 10.1088/1538-3873/aaecbe Blaum Hough, J. 2026, ebsbi v1.0.0: Neural simulation-based inference for eclipsing binaries, v1.0.0, Zenodo, doi: 10.5281/zenodo.19560120 Blaum Hough, J., Zhang, K., Bloom, J., van der Walt, S., & Cassese, B. 2026, Modified NBI (v0.4.1): Multi-moda...

    work page doi:10.1088/1538-3873/aaecbe 2019

  3. [3]

    , keywords =

    Chen, X., Wang, S., Deng, L., et al. 2020, ApJS, 249, 18, doi: 10.3847/1538-4365/ab9cae

    work page doi:10.3847/1538-4365/ab9cae 2020

  4. [4]

    E., Kochoska, A., Hey, D., et al

    Conroy, K. E., Kochoska, A., Hey, D., et al. 2020, ApJS, 250, 34, doi: 10.3847/1538-4365/abb4e2

    work page doi:10.3847/1538-4365/abb4e2 2020

  5. [5]

    2020, Proceedings of the National Academy of Science, 117, 30055, doi: 10.1073/pnas.1912789117

    Cranmer, K., Brehmer, J., & Louppe, G. 2020, Proceedings of the National Academy of Science, 117, 30055, doi: 10.1073/pnas.1912789117

    work page doi:10.1073/pnas.1912789117 2020

  6. [6]

    Goncalves, and Jakob H

    Deistler, M., Boelts, J., Steinbach, P., et al. 2025, arXiv e-prints, arXiv:2508.12939, doi: 10.48550/arXiv.2508.12939

    work page doi:10.48550/arxiv.2508.12939 2025

  7. [7]

    Eggleton, P. P. 1983, ApJ, 268, 368, doi: 10.1086/160960

    work page doi:10.1086/160960 1983

  8. [8]

    Fitzpatrick, E. L. 1999, PASP, 111, 63, doi: 10.1086/316293 Gaia Collaboration, Prusti, T., de Bruijne, J. H. J., et al. 2016, A&A, 595, A1, doi: 10.1051/0004-6361/201629272 Gaia Collaboration, Vallenari, A., Brown, A. G. A., et al. 2023, A&A, 674, A1, doi: 10.1051/0004-6361/202243940

    work page doi:10.1086/316293 1999

  9. [9]

    , keywords =

    Graham, M. J., Kulkarni, S. R., Bellm, E. C., et al. 2019, PASP, 131, 078001, doi: 10.1088/1538-3873/ab006c

    work page doi:10.1088/1538-3873/ab006c 2019

  10. [10]

    , year = 2019, month = dec, volume =

    Finkbeiner, D. 2019, ApJ, 887, 93, doi: 10.3847/1538-4357/ab5362

    work page doi:10.3847/1538-4357/ab5362 2019

  11. [11]

    Hermans, A

    Hermans, J., Delaunoy, A., Rozet, F., et al. 2021, arXiv e-prints, arXiv:2110.06581, doi: 10.48550/arXiv.2110.06581 Ivezi´ c,ˇZ., Kahn, S. M., Tyson, J. A., et al. 2019, ApJ, 873, 111, doi: 10.3847/1538-4357/ab042c SBI for Detached Eclipsing Binaries33

    work page doi:10.48550/arxiv.2110.06581 2021

  12. [12]

    2016, AJ, 151, 68, doi: 10.3847/0004-6256/151/3/68 18

    Kirk, B., Conroy, K., Prˇ sa, A., et al. 2016, AJ, 151, 68, doi: 10.3847/0004-6256/151/3/68

    work page doi:10.3847/0004-6256/151/3/68 2016

  13. [13]

    , keywords =

    Kochanek, C. S., Shappee, B. J., Stanek, K. Z., et al. 2017, PASP, 129, 104502, doi: 10.1088/1538-3873/aa80d9

    work page doi:10.1088/1538-3873/aa80d9 2017

  14. [14]

    Lueckmann, J

    Lueckmann, J.-M., Boelts, J., Greenberg, D. S., Gon¸ calves, P. J., & Macke, J. H. 2021, arXiv e-prints, arXiv:2101.04653, doi: 10.48550/arXiv.2101.04653

    work page doi:10.48550/arxiv.2101.04653 2021

  15. [15]

    Maxted, P. F. L. 2016, A&A, 591, A111, doi: 10.1051/0004-6361/201628579

    work page doi:10.1051/0004-6361/201628579 2016

  16. [16]

    , keywords =

    Miller, A. A., Abrams, N. S., Aldering, G., et al. 2025, PASP, 137, 094204, doi: 10.1088/1538-3873/ae02c5 Paczy´ nski, B., Szczygie l, D. M., Pilecki, B., & Pojma´ nski, G. 2006, MNRAS, 368, 1311, doi: 10.1111/j.1365-2966.2006.10223.x

    work page doi:10.1088/1538-3873/ae02c5 2025

  17. [17]

    Neural Density Estimation and Likelihood-free Inference , school =

    Papamakarios, G. 2019, arXiv e-prints, arXiv:1910.13233, doi: 10.48550/arXiv.1910.13233 Pietrzy´ nski, G., Graczyk, D., Gieren, W., et al. 2013, Nature, 495, 76, doi: 10.1038/nature11878 Prˇ sa, A., & Zwitter, T. 2005, ApJ, 628, 426, doi: 10.1086/430591 Prˇ sa, A., Batalha, N., Slawson, R. W., et al. 2011, AJ, 141, 83, doi: 10.1088/0004-6256/141/3/83 Prˇ ...

    work page doi:10.48550/arxiv.1910.13233 2019

  18. [18]

    M., Jayasinghe, T., Stanek, K

    Rowan, D. M., Jayasinghe, T., Stanek, K. Z., et al. 2022, MNRAS, 517, 2190, doi: 10.1093/mnras/stac2520

    work page doi:10.1093/mnras/stac2520 2022

  19. [19]

    , keywords =

    Shappee, B. J., Prieto, J. L., Grupe, D., et al. 2014, ApJ, 788, 48, doi: 10.1088/0004-637X/788/1/48 Soszy´ nski, I., Pawlak, M., Pietrukowicz, P., et al. 2016, AcA, 66, 405, doi: 10.48550/arXiv.1701.03105

    work page doi:10.1088/0004-637x/788/1/48 2014

  20. [20]

    M., Torres G., Zejda M., eds, Astronomical Society of the Pacific Conference Series Vol

    Southworth, J. 2015, in Astronomical Society of the Pacific Conference Series, Vol. 496, Living Together: Planets, Host Stars and Binaries, ed. S. M. Rucinski, G. Torres, & M. Zejda, 164, doi: 10.48550/arXiv.1411.1219

    work page doi:10.48550/arxiv.1411.1219 2015

  21. [21]

    2010, A&A Rv, 18, 67, doi: 10.1007/s00159-009-0025-1

    Torres, G., Andersen, J., & Gim´ enez, A. 2010, A&A Rv, 18, 67, doi: 10.1007/s00159-009-0025-1

    work page doi:10.1007/s00159-009-0025-1 2010

  22. [22]

    2019, MNRAS, 489, 1644, doi: 10.1093/mnras/stz2137

    Windemuth, D., Agol, E., Ali, A., & Kiefer, F. 2019, MNRAS, 489, 1644, doi: 10.1093/mnras/stz2137

    work page doi:10.1093/mnras/stz2137 2019

  23. [23]

    2023, in Machine Learning for Astrophysics, 38, doi: 10.48550/arXiv.2312.03824

    Zhang, K., Bloom, J., & Hernitschek, N. 2023, in Machine Learning for Astrophysics, 38, doi: 10.48550/arXiv.2312.03824

    work page doi:10.48550/arxiv.2312.03824 2023

  24. [24]

    S., Gaudi, B

    Zhang, K., Bloom, J. S., Gaudi, B. S., et al. 2021, AJ, 161, 262, doi: 10.3847/1538-3881/abf42e

    work page doi:10.3847/1538-3881/abf42e 2021