arxiv: 2604.12758 · v1 · submitted 2026-04-14 · 🌌 astro-ph.SR · astro-ph.EP· astro-ph.IM

Recognition: unknown

Variability classification of TESS targets in LOPS2, the first long-term pointing field of PLATO. Version 1 of the public variability catalogue

Mykyta Kliapets , Pablo Huijse , Jeroen Audenaert , Andrew Tkachenko , Marek Skarka , Paul F. X. Gregory , Dominic M. Bowman , Simon J. Murphy

show 23 more authors

Poojan Agrawal J\'ozsef M. Benk\H{o} Hannah Brinkman Nicholas Jannsen Yoshi Nike Emilia Eschen Allison Eto Dario J. Fritzewski Alex Kemp Viktor Khalack Gang Li Ricardo Ochoa-Armenta In\^es Rolo Nena Scheller Rose S. Stanley Keegan Thomson-Paressant Emese Plachy Vincent Vanlaer Mathijs Vanrespaille Jasmine Vrancken Haotian Wang Yian Xia George R. Ricker Conny Aerts

Authors on Pith no claims yet

Pith reviewed 2026-05-10 14:19 UTC · model grok-4.3

classification 🌌 astro-ph.SR astro-ph.EPastro-ph.IM

keywords variable starsTESS light curvesPLATO missionvariability classificationmachine learningLOPS2 fieldstellar pulsatorseclipsing binaries

0 comments

The pith

Machine learning on 38 million TESS light curves identifies 3.6 million candidate variable stars in PLATO's LOPS2 field.

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

The paper processes 38 million TESS light curves for 6 million stars in PLATO's first long-term pointing field using a deep neural network and a gradient-boosted decision tree ensemble. Their combined output flags roughly 72 percent of the curves as dominated by instrument or pipeline artifacts. The remaining 28 percent yields 3.6 million candidate variables that include pulsators, rotators, and eclipsing systems. A public catalogue is released so the community can study these stars ahead of PLATO's 2027 launch. The work matters because it supplies a ready-made variability list for the same sky region PLATO will monitor for years.

Core claim

We classified 38 million calibrated aperture light curves from the TESS-Gaia Light Curve pipeline for 6 million unique sources in LOPS2 with two machine learning frameworks -- a deep neural network and a feature-based gradient-boosted decision-tree ensemble. We combined their predictions to create this first version of the LOPS2 variability catalogue, performed manual vetting of a sub-sample of classified light curves, and a statistical analysis of the results to validate our methodology and to assess the variability properties and parameters of the stars in the catalogue. Our classification resulted in the identification of approximately 72% of the light curves having dominant instrument- 0

What carries the argument

Combined predictions from a deep neural network and a feature-based gradient-boosted decision-tree ensemble, followed by manual vetting of a subsample.

If this is right

Filtering candidates on colour, luminosity, dominant frequency, amplitude, and proximity of neighbours increases sample purity.
Candidate pulsators display a wide range of frequencies, amplitudes, rotation rates, and stellar parameters.
The released catalogue supplies one of the largest automated variability lists for immediate use by the community.
The same two-framework approach can be applied to future TESS sectors that overlap PLATO fields.

Where Pith is reading between the lines

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

The catalogue could serve as a target list for PLATO Guest Observer proposals focused on variable-star science.
Similar classification pipelines might be tested on upcoming wide-field surveys to handle even larger data volumes.
Discrepancies between the neural network and tree ensemble outputs could highlight specific artifact types worth separate study.

Load-bearing premise

The combined machine-learning predictions after manual vetting of a subsample reliably separate genuine stellar variability from TESS pipeline artifacts across the entire set of 38 million light curves.

What would settle it

Independent variability measurements from a different instrument or survey on a statistically significant random sample of the 3.6 million candidates would show whether the reported 28 percent fraction matches the true rate of detectable stellar variability.

Figures

Figures reproduced from arXiv: 2604.12758 by Alex Kemp, Allison Eto, Andrew Tkachenko, Conny Aerts, Dario J. Fritzewski, Dominic M. Bowman, Emese Plachy, Gang Li, George R. Ricker, Hannah Brinkman, Haotian Wang, In\^es Rolo, Jasmine Vrancken, Jeroen Audenaert, J\'ozsef M. Benk\H{o}, Keegan Thomson-Paressant, Marek Skarka, Mathijs Vanrespaille, Mykyta Kliapets, Nena Scheller, Nicholas Jannsen, Pablo Huijse, Paul F. X. Gregory, Poojan Agrawal, Ricardo Ochoa-Armenta, Rose S. Stanley, Simon J. Murphy, Viktor Khalack, Vincent Vanlaer, Yian Xia, Yoshi Nike Emilia Eschen.

**Figure 1.** Figure 1: Number of light curves for stars in LOPS2 classified per TESS sector. The entire bin height indicates all light curves for a given sector, where those with a dominant instrumental frequency f1 are indicated in purple and those with a second or third instrumental frequency (without double counting) in teal. The burgundy dotted line delineates classified nominal and extended mission data. Compared to Gregory… view at source ↗

**Figure 2.** Figure 2: Normalised distributions of the dominant frequency f1 (top) and frequency density of the entire frequency space (bottom) before (purple) and after (teal) downsampling the cadence with a factor 3 for extendedmission p-mode pulsators. Histograms are plotted from 1 000 samples for each candidate frequency within the uncertainty range. Kernel Density Estimators are plotted as full lines and were computed dir… view at source ↗

**Figure 3.** Figure 3: Light curves (grey) and dimensionless Lomb Scargle periodograms as defined in VanderPlas (2018) overplotted in orange for Gaia DR3 5317171818865844864 predicted as CONTACT_ROT by the neural network (0.85) and INSTRUMENT by XGBoost (0.55) from Sector 10 (upper panel) and in green for Gaia DR3 4775678010208086016 predicted as DSCT_BCEP by the neural network (0.89) and INSTRUMENT by XGBoost (0.90) from Sect… view at source ↗

**Figure 4.** Figure 4: The ensemble confusion matrix normalised by row. The first number in each cell is a fraction of the true label that is retrieved. The second and third rows are changes from using just the deep learning classifier and XGBoost, respectively. A positive delta (second and third row) is better on the main diagonal, and a negative delta is better off the main diagonal. 4. Navigating the catalogue 4.1. Thresholdi… view at source ↗

**Figure 5.** Figure 5: Same as [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: Normalised distributions of dominant frequencies f1 for classified light curves (in colour) with final scores above 0.95 (top), 0.75 (middle), and 0.5 (bottom) and those of the training set (black outline). If a single Gaia DR3 had multiple light curves, all of them were included. 0.5 1.0 1.5 Period [d] 0 5000 10000 15000 20000 25000 30000 Star Index GDOR_SPB 5 10 15 20 Frequency [1/d] 0 20000 40000 60000 … view at source ↗

**Figure 7.** Figure 7: Stacked amplitude spectra for candidate g-mode (in period, top panel) and p-mode (in frequency, bottom panel) pulsators. Only candidate light curves with final scores above 0.95 are plotted for visibility purposes. Hey & Aerts 2024). The lower l = 2 mode ridge (Li et al. 2020; Hey & Aerts 2024) and the upper r-mode ridge (Li et al. 2020), both with lower amplitudes than for f1, can also be clearly seen. T… view at source ↗

**Figure 8.** Figure 8: Normalised distributions of the dominant period 1/ f1 for the training set (purple) and classified light curves with final scores above 0.5 (teal). Due to small 1/ f1 errors, both histograms and Kernel Density Estimators (as full lines) were computed directly from the point estimates. 0 1 2 3 4 frot [d 1 ] Norm. Histogram [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗

**Figure 10.** Figure 10: HRD with a sub-sample of light curves classified as CONTACT_ROT (orange crosses), DSCT_BCEP (green), GDOR_SPB (magenta), RRLYR_CEPH (yellow stars), and SOLARLIKE (light blue). Vertical ridges are due to Gaia DR3 temperature grid systematics. to the training set distributions (in black). For some classes, the distributions are shifted towards cooler temperatures, which on a case-by-case basis is either a… view at source ↗

**Figure 9.** Figure 9: Normalised distributions of the near-core rotation frequency frot computed from the recipe in Aerts et al. (2025) for the training set (purple) and for classified light curves with final scores above 0.5 (teal). Histograms are plotted from 1 000 samples for each candidate frequency within the uncertainty range. Kernel Density Estimators are plotted as full lines and were computed directly from the point … view at source ↗

read the original abstract

The PLAnetary Transits and Oscillations of stars (PLATO) mission is expected to launch in January 2027. A total of 8\% of its data rate will be dedicated to complementary science targets selected from approved Guest Observer proposals. We seek to provide an open-source catalogue of variable stars in PLATO's first long-term observing field, LOPS2. We want to use existing observations from the Transiting Exoplanet Survey Satellite (TESS), which has observed many stars in LOPS2. We classified 38 million calibrated aperture light curves from the TESS-Gaia Light Curve pipeline (TGLC, $G\lesssim17$) for 6 million unique sources in LOPS2 with two machine learning frameworks -- a deep neural network and a feature-based gradient-boosted decision-tree ensemble. We combined their predictions to create this first version of the LOPS2 variability catalogue, performed manual vetting of a sub-sample classified light curves, and a statistical analysis of the results to validate our methodology and to assess the variability properties and parameters of the stars in the catalogue. Our classification resulted in the identification of approximately 72% of the light curves having dominant instrument- or pipeline-induced signal, with the remaining 28% representing 3.6 million individual candidate variable stars, including pulsating, rotating, and eclipsing stars. Candidate pulsators exhibit varied behaviour in terms of their frequencies, amplitudes, rotation, and fundamental parameters. To ensure purity of the samples, filtering on colour, luminosity, the dominant frequency and its amplitude, and presence of close neighbours is helpful. We provide the first version of our PLATO LOPS2 variability catalogue to the community for further study and scrutiny. It is to date one of the largest catalogues of variable stars from an automated classification pipeline.

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.

Desk Editor's Note private letter to a colleague

This paper releases a large public catalog of 3.6 million variable star candidates in the PLATO LOPS2 field from TESS data, using combined DNN and GBT classification on 38 million light curves.

read the letter

The core deliverable is a new variability catalog for the LOPS2 field, built by classifying TESS-Gaia light curves for six million sources. They flag 72 percent as pipeline artifacts and keep the rest as candidates across pulsators, rotators, and eclipsing systems, then add some manual checks on a subsample and basic statistics on frequencies and amplitudes. The work is directly useful for PLATO complementary science and anyone needing a starting list in that sky patch. Processing that volume and releasing the data openly is the practical strength here; the methods are standard but applied at scale to a mission-relevant field. The soft spot is the classification reliability. The 28 percent figure depends on the ensemble generalizing across the full set, yet the description relies on training details, a vetted subsample, and statistical summaries without reported precision, recall, or agreement metrics on a held-out set that covers the range of TESS systematics. Modest per-class errors could shift the candidate count by hundreds of thousands given the imbalance, so the raw numbers would be stronger with uncertainty estimates or cross-validation results. Filters on color, luminosity, and neighbors are noted as helpful for purity, which is reasonable. This is for PLATO planners and stellar variability researchers who want to cross-match or select targets from an existing large list. A reader focused on data products rather than new theory will get the most out of it. The paper is coherent and the catalog substantial enough to deserve referee time, even if the validation section needs tightening.

Referee Report

2 major / 3 minor

Summary. The manuscript describes the creation of the first version of a public variability catalogue for PLATO's LOPS2 field. Using 38 million TESS-Gaia Light Curve (TGLC) aperture light curves for 6 million sources, the authors apply two machine-learning frameworks—a deep neural network and a feature-based gradient-boosted decision-tree ensemble—combine their predictions, perform manual vetting on a sub-sample, and conduct statistical analysis. This yields a classification in which ~72% of light curves are dominated by instrument- or pipeline-induced signals and the remaining 28% (~3.6 million candidates) are flagged as pulsating, rotating, or eclipsing variables. The catalogue is released publicly with suggestions for purity filters based on colour, luminosity, frequency, amplitude, and neighbours.

Significance. If the reported classification fractions prove robust, the work would deliver one of the largest public variability catalogues derived from an automated pipeline, directly supporting complementary-science target selection for PLATO's first long-term field. The combination of two independent ML frameworks, manual vetting, and statistical checks is a constructive approach, and the public release plus explicit purity-filter recommendations add practical value. The significance is currently limited by the absence of quantitative performance metrics that would allow readers to gauge uncertainty in the headline 72%/28% split.

major comments (2)

[Section 3 (Machine Learning Classification and Validation)] The headline result that 28% of the 38 million light curves are genuine variable-star candidates (and thus 3.6 million objects) rests on the assumption that the combined DNN + GBT predictions generalize reliably to the full TGLC set. The manuscript describes model training, output combination, sub-sample manual vetting, and statistical checks, but supplies no precision, recall, confusion matrix, or agreement statistics on a large, representative held-out test set that spans the observed range of TESS systematics (scattered light, momentum dumps, etc.) and the full diversity of variable classes. Without these numbers, even modest per-class error rates under the reported class imbalance can shift the reported fractions by hundreds of thousands of objects.
[Section 4 (Results)] The generalization step from the manually vetted sub-sample to the entire 38 million light curves is not accompanied by any quantitative uncertainty estimate. The central claim of the catalogue therefore lacks the error bars or sensitivity analysis that would be required to assess how robust the 72%/28% division is to plausible variations in model performance.

minor comments (3)

[Abstract] The abstract quotes approximate percentages and an integer count (3.6 million); providing the exact counts or ranges with any available uncertainty would improve precision.
[Figure captions] Several example light-curve figures would benefit from explicit labels indicating the final assigned variability class and the dominant frequency/amplitude values used in the statistical analysis.
[Section 3] A short table summarizing the exact training/validation split sizes, hyper-parameter choices, and any agreement metric between the DNN and GBT outputs would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's comments highlighting the need for more rigorous quantitative validation of our machine learning classifications. We have revised the manuscript to include additional performance metrics and uncertainty analyses as detailed in the point-by-point responses below.

read point-by-point responses

Referee: [Section 3 (Machine Learning Classification and Validation)] The headline result that 28% of the 38 million light curves are genuine variable-star candidates (and thus 3.6 million objects) rests on the assumption that the combined DNN + GBT predictions generalize reliably to the full TGLC set. The manuscript describes model training, output combination, sub-sample manual vetting, and statistical checks, but supplies no precision, recall, confusion matrix, or agreement statistics on a large, representative held-out test set that spans the observed range of TESS systematics (scattered light, momentum dumps, etc.) and the full diversity of variable classes. Without these numbers, even modest per-class error rates under the reported class imbalance can shift the reported fractions by hundreds of thousands of objects.

Authors: We acknowledge this limitation in the current version of the manuscript. While we performed manual vetting on a sub-sample and conducted statistical checks, we did not include a comprehensive held-out test set evaluation spanning all systematics. In the revised manuscript, we will add precision, recall, and a confusion matrix derived from the cross-validation during model training, as well as the agreement statistics between the DNN and GBT on the full dataset. We will also discuss the challenges in creating a fully representative test set for TESS data. These additions will help quantify the potential impact of misclassifications on the reported fractions. revision: yes
Referee: [Section 4 (Results)] The generalization step from the manually vetted sub-sample to the entire 38 million light curves is not accompanied by any quantitative uncertainty estimate. The central claim of the catalogue therefore lacks the error bars or sensitivity analysis that would be required to assess how robust the 72%/28% division is to plausible variations in model performance.

Authors: We agree that the manuscript would be strengthened by quantitative uncertainty estimates. In the revised version, we will include a sensitivity analysis varying the model combination parameters and report the resulting variation in the 28% fraction. We will also provide uncertainty estimates based on the vetted sub-sample proportions and discuss potential biases from the class imbalance. This will allow readers to better gauge the robustness of the headline results. revision: yes

Circularity Check

0 steps flagged

No significant circularity; classification applies standard ML models to external TESS data.

full rationale

The paper trains a DNN and GBT ensemble on variability patterns from TESS light curves, combines outputs, performs manual vetting on a sub-sample, and applies the result to the full 38M set to report the 72%/28% split. No step reduces a claimed prediction or uniqueness result to a fitted parameter or self-citation by construction; the output fractions are direct consequences of the trained classifiers on independent observations rather than a redefinition or tautological renaming of inputs. The pipeline remains self-contained against external benchmarks with no load-bearing self-referential definitions or ansatz smuggling.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the quality of TESS input data and the generalization ability of the two ML models; no new physical entities are introduced.

free parameters (1)

ML ensemble combination rules
Rules for merging DNN and gradient-boosted tree predictions are not detailed and implicitly tuned during model development.

axioms (2)

domain assumption TGLC provides calibrated aperture light curves that faithfully capture stellar signals after removal of instrumental effects
Directly used as the input for classifying 38 million light curves from 6 million sources.
domain assumption Models trained on known variable stars generalize to classify variability in new TESS observations of the LOPS2 field
Foundation for applying the deep neural network and feature-based ensemble to the full dataset.

pith-pipeline@v0.9.0 · 5811 in / 1455 out tokens · 65358 ms · 2026-05-10T14:19:41.671568+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Plato's view on supermassive black hole binaries: Exploring the faint limit of ESA's Plato space mission
astro-ph.GA 2026-05 unverdicted novelty 5.0

Simulations show Plato can recover relativistic photometric signatures of supermassive black hole binaries in bright quasars (G≤18) via Bayesian inference on mock light curves.

Reference graph

Works this paper leans on

110 extracted references · 9 canonical work pages · cited by 1 Pith paper · 8 internal anchors

[1]

2021, Rev

Aerts, C. 2021, Rev. Mod. Phys, 93, 015001

2021
[2]

2025, A&A, 704, A332

Aerts, C. 2025, A&A, 704, A332

2025
[3]

Aerts, C., Christensen-Dalsgaard, J., & Kurtz, D. W. 2010, Asteroseismology (Springer Science & Business Media)

2010
[4]

2023, A&A, 672, A183

Aerts, C., Molenberghs, G., & De Ridder, J. 2023, A&A, 672, A183

2023
[5]

& Tkachenko, A

Aerts, C. & Tkachenko, A. 2024, A&A, 692, R1

2024
[6]

& Tkachenko, A

Aerts, C. & Tkachenko, A. 2024, in 8th TESS/15th Kepler Asteroseismic Sci- ence Consortium Workshop, 22

2024
[7]

S., & Hey, D

Aerts, C., Van Reeth, T., Mombarg, J. S., & Hey, D. 2025, A&A, 695, A214

2025
[8]

2025, A&A, 696, A111

Antoci, V ., Cantiello, M., Khalack, V ., et al. 2025, A&A, 696, A111

2025
[9]

M., Lim, P

Astropy, C., Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ, 935

2022
[10]

2025, Ap&SS, 370, 72 Article number, page 12 of 18 M

Audenaert, J. 2025, Ap&SS, 370, 72 Article number, page 12 of 18 M. Kliapets et al.: PLATO LOPS2 Variability Catalogue, Version1

2025
[11]

S., Handberg, R., et al

Audenaert, J., Kuszlewicz, J. S., Handberg, R., et al. 2021, AJ, 162, 209

2021
[12]

Audenaert, J., Muthukrishna, D., Gregory, P. F. X., Hogg, D. W., & Villar, V . A. 2025, in 1st ICML Workshop on Foundation Models for Structured Data

2025
[13]

& Tkachenko, A

Audenaert, J. & Tkachenko, A. 2022, A&A, 666, A76

2022
[14]

Audenaert, J., Tkachenko, A., Skarka, M., Eschen, Y . N. E., & Muthukrishna, D. 2024, in 8th TESS/15th Kepler Asteroseismic Science Consortium Workshop, 47

2024
[15]

R., Murphy, S

Barac, N., Bedding, T. R., Murphy, S. J., & Hey, D. R. 2022, MNRAS, 516, 2080

2022
[16]

S., Sahoo, S

Baran, A. S., Sahoo, S. K., Sanjayan, S., & Ostrowski, J. 2021, MNRAS, 503, 3828

2021
[17]

H., Bedding, T

Barbara, N. H., Bedding, T. R., Fulcher, B. D., Murphy, S. J., & Van Reeth, T. 2022, MNRAS, 514, 2793

2022
[18]

R., Murphy, S

Bedding, T. R., Murphy, S. J., Crawford, C., et al. 2023, ApJ, 946, L10

2023
[19]

J., Koch, D., Basri, G., et al

Borucki, W. J., Koch, D., Basri, G., et al. 2010, Science, 327, 977

2010
[20]

Bowman, D. M. & Bugnet, L. 2026, in Encyclopedia of Astrophysics (First Edi- tion), first edition edn., ed. I. Mandel (Oxford: Elsevier), 133–153

2026
[21]

M., Kurtz, D

Bowman, D. M., Kurtz, D. W., Breger, M., Murphy, S. J., & Holdsworth, D. L. 2016, MNRAS, 460, 1970

2016
[22]

2001, Mach

Breiman, L. 2001, Mach. Learn., 45, 5

2001
[23]

M., Michielsen, M., et al

Burssens, S., Bowman, D. M., Michielsen, M., et al. 2023, Nat. Astron., 7, 913

2023
[24]

M., et al

Buysschaert, B., Aerts, C., Bowman, D. M., et al. 2018, A&A, 616, A148

2018
[25]

2018, Astron

Cabral, J., Sánchez, B., Ramos, F., et al. 2018, Astron. Comput

2018
[26]

Assessment of PLATO Science Performance

Cabrera, J., Rauer, H., Samadi, R., et al. 2026, Exp. Astron., submitted, arXiv:2604.04818

work page internal anchor Pith review Pith/arXiv arXiv 2026
[27]

A., Tenenbaum, P., Twicken, J

Caldwell, D. A., Tenenbaum, P., Twicken, J. D., et al. 2020, RNAAS, 4, 201

2020
[28]

2019, R version, 90, 40

Chen, T., He, T., Benesty, M., & Khotilovich, V . 2019, R version, 90, 40

2019
[29]

2015, R package version 0.4-2, 1, 1

Chen, T., He, T., Benesty, M., et al. 2015, R package version 0.4-2, 1, 1

2015
[30]

L., Angus, R., David, T., et al

Colman, I. L., Angus, R., David, T., et al. 2024, AJ, 167, 189

2024
[31]

2007, A&A, 475, 1159

Debosscher, J., Sarro, L., Aerts, C., et al. 2007, A&A, 475, 1159

2007
[32]

& Alonso, R

Deeg, H. & Alonso, R. 2024, Contrib. Astron. Obs. Skalnaté Pleso, 54, 142

2024
[33]

2020, Front

Dong, X., Yu, Z., Cao, W., Shi, Y ., & Ma, Q. 2020, Front. Comput. Sci., 14, 241

2020
[34]

2004, A&A, 414, L17

Dupret, M.-A., Grigahcène, A., Garrido, R., Gabriel, M., & Scuflaire, R. 2004, A&A, 414, L17

2004
[35]

Eschen, Y . N. E., Bayliss, D., Wilson, T. G., et al. 2024, MNRAS, 535, 1778

2024
[36]

& Mowlavi, N

Eyer, L. & Mowlavi, N. 2008, in Journal of Physics Conference Series, V ol. 118, Journal of Physics Conference Series (IOP), 012010

2008
[37]

Friedman, J. H. 2001, Ann. Stat., 1189

2001
[38]

2024, A&A, 681, A13

Fritzewski, D., Van Reeth, T., Aerts, C., et al. 2024, A&A, 681, A13

2024
[39]

2025, A&A, 698, A253

Fritzewski, D., Vanrespaille, M., Aerts, C., et al. 2025, A&A, 698, A253

2025
[40]

J., Kemp, A., Li, G., & Aerts, C

Fritzewski, D. J., Kemp, A., Li, G., & Aerts, C. 2026, A&A, 706, A131 Gaia Collaboration; De Ridder, J., Ripepi, V ., Aerts, C., et al. 2023, A&A, 674, A36 Gaia Collaboration; Prusti, T., De Bruijne, J., Brown, A. G., et al. 2016, A&A, 595, A1

2026
[41]

2020, Nat

Geirhos, R., Jacobsen, J.-H., Michaelis, C., et al. 2020, Nat. Mach. Intell., 2, 665

2020
[42]

Gregory, P. F. X., Audenaert, J., Kliapets, M., et al. 2026, ApJ, submitted, arXiv:2604.07437 Grigahcène, A., Antoci, V ., Balona, L., et al. 2010, ApJ, 713, L192

work page internal anchor Pith review Pith/arXiv arXiv 2026
[43]

Guo, C., Pleiss, G., Sun, Y ., & Weinberger, K. Q. 2017, in International confer- ence on machine learning, PMLR, 1321–1330

2017
[44]

2013, EAS Publ

Hambleton, K., Degroote, P., Conroy, K., et al. 2013, EAS Publ. Ser., 64, 285

2013
[45]

& Brandt, T

Han, T. & Brandt, T. D. 2023, AJ, 165, 71

2023
[46]

N., White, T

Handberg, R., Lund, M. N., White, T. R., et al. 2021, AJ, 162, 170

2021
[47]

2025, AJ, 170, 15

Hattori, S., Angus, R., Foreman-Mackey, D., Lu, Y ., & Colman, I. 2025, AJ, 170, 15

2025
[48]

2009, ARA&A, 47, 211

Heber, U. 2009, ARA&A, 47, 211

2009
[49]

PLATO input catalogs for technical calibration and fine guidance

Heller, R., Jiang, C., Bluhm, P., et al. 2026, Exp. Astron., submitted, arXiv:2604.02437

work page internal anchor Pith review Pith/arXiv arXiv 2026
[50]

& Aerts, C

Hey, D. & Aerts, C. 2024, A&A, 688, A93

2024
[51]

R., Montet, B

Hey, D. R., Montet, B. T., Pope, B. J., Murphy, S. J., & Bedding, T. R. 2021, AJ, 162, 204

2021
[52]

Higgins, M. E. & Bell, K. J. 2023, AJ, 165, 141

2023
[53]

2025, A&A, 701, A150

Huijse, P., De Ridder, J., Eyer, L., et al. 2025, A&A, 701, A150

2025
[54]

2024, A&A, 681, A18

Jannsen, N., De Ridder, J., Seynaeve, D., et al. 2024, A&A, 681, A18

2024
[55]

2025, A&A, 694, A185

Jannsen, N., Tkachenko, A., Royer, P., et al. 2025, A&A, 694, A185

2025
[56]

S., et al

Kemp, A., Vrancken, J., Mombarg, J. S., et al. 2025, A&A, 704, A280

2025
[57]

2025, A&A, 703, A240

Kliapets, M., Huijse, P., Tkachenko, A., et al. 2025, A&A, 703, A240

2025
[58]

A., Rix, H.-W., Aerts, C., et al

Kollmeier, J. A., Rix, H.-W., Aerts, C., et al. 2026, AJ, 171, 52

2026
[59]

2021, RNAAS, 5, 234

Kunimoto, M., Huang, C., Tey, E., et al. 2021, RNAAS, 5, 234

2021
[60]

2017, NeurIPS, 30

Lakshminarayanan, B., Pritzel, A., & Blundell, C. 2017, NeurIPS, 30

2017
[61]

M., & Van Reeth, T

Lecoanet, D., Bowman, D. M., & Van Reeth, T. 2022, MNRASL, 512, L16

2022
[62]

R., et al

Li, G., Van Reeth, T., Bedding, T. R., et al. 2020, MNRAS, 491, 3586

2020
[63]

N., Handberg, R., Buzasi, D

Lund, M. N., Handberg, R., Buzasi, D. L., et al. 2021, ApJS, 257, 53

2021
[64]

2026, in ICLR 2026 Workshop on Foundation Models for Science: Real-World Impact and Science-First Design

Mercader-Perez, P., Cuesta-Lazaro, C., Muthukrishna, D., et al. 2026, in ICLR 2026 Workshop on Foundation Models for Science: Real-World Impact and Science-First Design

2026
[65]

Mitchell, T. M. 1980, Technical Report No. CBM-TR-117

1980
[66]

Mombarg, J. S. G., Aerts, C., Van Reeth, T., & Hey, D. 2024, A&A, 691, A131

2024
[67]

M., et al

Montalto, M., Piotto, G., Marrese, P. M., et al. 2021, A&A, 653, A98

2021
[68]

The PLATO Input Catalogue of targets (tPIC) for the first Long Pointing Field

Montalto, M., Piotto, G., Marrese, P. M., et al. 2026, A&A, submitted, arXiv:2604.03369

work page internal anchor Pith review Pith/arXiv arXiv 2026
[69]

J., Hey, D., Van Reeth, T., & Bedding, T

Murphy, S. J., Hey, D., Van Reeth, T., & Bedding, T. R. 2019, MNRAS, 485, 2380

2019
[70]

2022, A&A, 658, A31

Nascimbeni, V ., Piotto, G., Börner, A., et al. 2022, A&A, 658, A31

2022
[71]

2025, A&A, 694, A313

Nascimbeni, V ., Piotto, G., Cabrera, J., et al. 2025, A&A, 694, A313

2025
[72]

The PLATO field selection process III. Selection of the Prime Sample for the LOPS2 field

Nascimbeni, V ., Piotto, G., Granata, V ., et al. 2026, A&A, submitted, arXiv:2604.03365 Pápics, P. I., Briquet, M., Baglin, A., et al. 2012, A&A, 542, A55

work page internal anchor Pith review Pith/arXiv arXiv 2026
[73]

2018, in International conference on in- telligent data communication technologies and internet of things, Springer, 758–763

Parmar, A., Katariya, R., & Patel, V . 2018, in International conference on in- telligent data communication technologies and internet of things, Springer, 758–763

2018
[74]

G., Aerts, C., Pápics, P

Pedersen, M. G., Aerts, C., Pápics, P. I., et al. 2021, Nat. Astron., 5, 715

2021
[75]

Pedersen, M. G. & Bell, K. J. 2023, AJ, 165, 239

2023
[76]

2021, in International Conference on Space Optics—ICSO 2020, V ol

Pertenais, M., Cabrera, J., Paproth, C., et al. 2021, in International Conference on Space Optics—ICSO 2020, V ol. 11852, SPIE, 2043–2054

2021
[77]

2026, RNAAS, 10, 69

Petitpas, G., Haviland, J., Han, T., et al. 2026, RNAAS, 10, 69

2026
[78]

2019, in Stars and their Variability Observed from Space, 465–470

Plachy, E. 2019, in Stars and their Variability Observed from Space, 465–470

2019
[79]

2021, ApJS, 253, 11

Plachy, E., Pál, A., Bódi, A., et al. 2021, ApJS, 253, 11

2021
[80]

& Szabó, R

Plachy, E. & Szabó, R. 2021, Front. Astron. Space Sci., 7, 577695

2021

Showing first 80 references.