pith. machine review for the scientific record. sign in

arxiv: 2604.07437 · v1 · submitted 2026-04-08 · 🌌 astro-ph.IM · astro-ph.SR

Recognition: no theorem link

ASTRAFier: A Novel and Scalable Transformer-based Stellar Variability Classifier

Paul F. X. Gregory , Jeroen Audenaert , Mykyta Kliapets , Daniel Muthukrishna , Andrew Tkachenko , Marek Skarka , Marc Hon , George R. Ricker
Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:07 UTC · model grok-4.3

classification 🌌 astro-ph.IM astro-ph.SR
keywords stellar variability classificationtransformer modellight curve time seriesKepler missionTESS missionmachine learningend-to-end classification
0
0 comments X

The pith

ASTRAFier classifies stellar variability types directly from light curves using a hybrid Transformer model.

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

The paper introduces ASTRAFier as an end-to-end neural network system that sorts photometric time series into categories of stellar variability. It combines transformer attention with bidirectional LSTM layers and convolutional networks so that classification happens on the raw light curve data without any hand-crafted features. The model is trained and tested on Kepler and TESS observations, reaching 94.26 percent accuracy on Kepler data and 88.22 percent on TESS data. It is then run on nearly three million TESS light curves from specific sectors to generate a public variability catalog. This matters because current and upcoming space surveys produce far more light curves than traditional methods can process efficiently.

Core claim

ASTRAFier integrates Transformer, BiLSTM, and CNN components into a single model that classifies stellar variability classes from raw time series light curves, achieving 94.26 percent accuracy on Kepler data and 88.22 percent on TESS data while scaling to process approximately 2.8 million TESS observations and release the resulting catalog.

What carries the argument

The ASTRAFier architecture, which fuses Transformer attention mechanisms with bidirectional LSTM for long-range sequence dependencies and CNN layers for local feature detection to enable direct end-to-end classification from light curve time series.

If this is right

  • Large photometric datasets can be classified without the time-consuming step of manual feature engineering.
  • New catalogs of stellar variability become feasible to produce at the scale of millions of light curves.
  • The framework supports ongoing updates as additional TESS sectors or future survey data arrive.

Where Pith is reading between the lines

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

  • The same architecture could be retrained on combined data from multiple surveys to reduce instrument-specific biases in variability labels.
  • Population-level statistics derived from the released catalog might reveal previously hidden trends in the occurrence rates of different variability types.
  • Similar hybrid models could be tested on other time-domain astronomy tasks such as transient detection or exoplanet signal identification.

Load-bearing premise

The labels used to train and validate the model accurately represent distinct stellar variability classes and the model generalizes to new TESS data without large instrument-specific biases or overfitting.

What would settle it

A comparison of ASTRAFier outputs against independently verified variability labels on a held-out sample of several thousand TESS light curves would directly test whether the reported accuracies are reproducible.

Figures

Figures reproduced from arXiv: 2604.07437 by Andrew Tkachenko, Daniel Muthukrishna, George R. Ricker, Jeroen Audenaert, Marc Hon, Marek Skarka, Mykyta Kliapets, Paul F. X. Gregory.

Figure 1
Figure 1. Figure 1: A Transformer encoder layer. Figure reproduced from Vaswani et al. (2017). 2.2. Long Short-Term Memory (LSTM) The foundation for LSTMs (Hochreiter & Schmid￾huber 1997) was laid by Recurrent Neural Networks (RNNs). Unlike feedforward neural networks, RNNs are designed to process sequences of data by maintaining a hidden state that evolves over time. At each time step t, the network updates its hidden state … view at source ↗
Figure 3
Figure 3. Figure 3: A 1-D kernel of size 3. This kernel slides along the input sequence stride steps at a time, producing a new sequence through a multiplication of its learned weights and the input sequence. While this example shows a CNN limited to handling 1-channel inputs and 1-channel outputs, we can general￾ize to handle multi-channel inputs and outputs as well. To handle a multi-channel input, we use a multi-channel ke… view at source ↗
Figure 2
Figure 2. Figure 2: An LSTM block at time step t. xt is index t of the input sequence, ct is the cell state at time t, and ht is the hidden state at time t. An LSTM module consists of many such blocks, typically one for each time step in the input se￾quence. The LSTM outputs its hidden states [h1, h2, ..., hT ]. In our model, we make use of a bidirectional LSTM (BiLSTM, Schuster & Paliwal 1997). This expands on the LSTM by pr… view at source ↗
Figure 4
Figure 4. Figure 4: A visualization of a 1-D convolution with 3 input channels and 2 output channels. function has been found to improve performance when modeling sequential data (Dauphin et al. 2016). 3. MODEL ARCHITECTURE Recent research has shown the advantages of integrat￾ing attention mechanisms, CNNs, and LSTMs due to their complementary capabilities in handling sequential data (e.g., Shen et al. 2024; Zhang et al. 2023… view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of the ASTRAFier model. The gray box highlights a single block, which is stacked three times. Each block contains a BiLSTM (with a CNN Projec￾tion), a Transformer encoder, and a CNN module. Add and Norm refers to a residual (skip) connection followed by layer normalization, which facilitates gradient flow and stabilizes training. Positional information is injected into the initial embeddings u… view at source ↗
Figure 6
Figure 6. Figure 6: A preprocessed light curve is embedded into a higher dimension. demonstrated that the post BiLSTM convolutional pro￾jection approach yielded higher classification accuracy, likely because the learned kernels better preserve the salient features extracted by the BiLSTM passes. While this approach increases the computational complexity and parameter count, the gain in predictive performance justifies the add… view at source ↗
Figure 7
Figure 7. Figure 7: The CNN module. as: z = PT t=1 mt · ht max  1, PT t=1 mt  (6) where ht is the t-th row of H. The pooled embed￾ding is then passed through a 3-layer MLP for classi￾fication. The MLP consists of a linear layer projecting from 64 to 128 dimensions, followed by Layer Normaliza￾tion (LayerNorm, Ba et al. 2016), SiLU activation, and dropout (p=0.2). A second linear layer reduces the di￾mension from 128 to 32, … view at source ↗
Figure 8
Figure 8. Figure 8: The confusion matrix on the Kepler holdout set for the model trained only on Kepler data. known challenge in variability classification (e.g., Au￾denaert et al. 2021; Barbara et al. 2022; Hey & Aerts 2024). Our model also fails to correctly identify the RRLYR CEPH class more often with most of the confu￾sion being again with the CONTACT ROT class. How￾ever, there are only 12 RRLYR CEPH stars in our hold￾ou… view at source ↗
Figure 9
Figure 9. Figure 9: The confusion matrix on the TESS holdout set for the model trained only on TESS data. where less obvious or noisier eclipses will present more challenging cases. 6.3. TESS and Kepler Using the refined class structure, we trained our fi￾nal deployment model on a combined dataset of TESS (Sect. 4.2) and Kepler (Sect. 4.1) light curves, remov￾ing any Kepler targets already present in the TESS set to avoid dat… view at source ↗
Figure 10
Figure 10. Figure 10: The confusion matrix on the TESS holdout set for the final model trained on both Kepler and TESS data [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The UMAP reduction of the data points in our final QLP holdout set extracted before the final MLP layer. analyzing the astrophysical properties exhibited by sub￾populations assigned a certain class, and support this with detailed inspections of their light curves and am￾plitude spectra. In [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Normalized distributions of effective temperatures (top), dominant variability (middle), and its amplitude (bottom) of the labeled set (black outline) and classified targets from TESS Sectors 14, 15, and 26 with probabilities higher than 0.5 (color) and 0.8 (hash). If a star had more than one light curve in the Kepler field of view, only the one with the higher probability was plotted. Distributions have … view at source ↗
Figure 13
Figure 13. Figure 13: Stacked amplitude spectra of candidate g￾mode (top, in period) and p-mode pulsators (bottom, in fre￾quency), for which the prediction probability is higher than 0.5. Stars of each of the two classes are sorted by the domi￾nant variability. time-steps. While this could lead to a more precise light curve with more distinct variability, especially for p￾mode pulsators, it is possible that the longer sequence… view at source ↗
Figure 14
Figure 14. Figure 14: HR Diagram of randomly-sampled high￾probability candidate light curves for all classes except DSCT BCEP and GDOR SPB (top panel) and, separately, only DSCT BCEP and GDOR SPB with normalized proba￾bilities for the two classes (bottom panel). Black outlines mark stars with the secondary class having a normalized probability higher than 0.2 — potentially hybrid pulsators. A vertical line at 15,000 K is a Gai… view at source ↗
read the original abstract

Photometric missions such as Kepler and TESS have generated millions of light curves covering almost the entire sky, offering unprecedented opportunities to study stellar variability and advance our understanding of the Universe. In this data-rich environment, machine learning has emerged as a powerful tool to efficiently and accurately process and classify light curves according to their type of stellar variability. In this work, we introduce ASTRAFier: a novel Transformer-based model for variability classification that integrates Bidirectional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNNs). The model operates directly on time series without requiring feature engineering, creating an easy-to-maintain and efficient end-to-end classification framework. We train and validate our model using both Kepler and TESS light curves and, respectively, achieve a classification accuracy of $94.26\%$ on Kepler and $88.22\%$ on TESS. We demonstrate scalability by deploying our model on $\sim 2.8$ million TESS light curves from sectors 14, 15, and 26 (Kepler Field-of-View) delivered by MIT's Quick-look Pipeline (QLP) and release the resulting stellar variability catalog.

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 manuscript introduces ASTRAFier, a Transformer-based model that integrates BiLSTM and CNN components to classify stellar variability directly from Kepler and TESS photometric time series without feature engineering. It reports classification accuracies of 94.26% on Kepler data and 88.22% on TESS data, demonstrates scalability by applying the model to approximately 2.8 million TESS light curves from sectors 14, 15, and 26, and releases the resulting stellar variability catalog.

Significance. If the performance claims are supported by rigorous validation, the end-to-end framework could offer a maintainable approach for processing large photometric datasets from ongoing and future surveys. The public release of the catalog derived from 2.8 million light curves constitutes a concrete community resource that strengthens the work's potential utility.

major comments (2)
  1. [Abstract] Abstract: The stated accuracies (94.26% Kepler, 88.22% TESS) are presented without any information on dataset splits, class balance, cross-validation strategy, preprocessing, or error analysis. These omissions directly affect evaluation of the central performance claims.
  2. [Results/Deployment] Deployment and results sections: The application to ~2.8 million TESS light curves assumes cross-instrument generalization, yet no cross-mission hold-out experiments, label-source audits, or ablation studies addressing domain shift (differences in cadence, noise, and precision between Kepler and TESS) are described.
minor comments (2)
  1. [Abstract/Methods] The abstract and methods would benefit from an explicit architectural diagram or pseudocode clarifying how the Transformer, BiLSTM, and CNN components are combined.
  2. [Introduction] Ensure consistent definition of acronyms (BiLSTM, CNN, QLP) on first use and provide a brief comparison to prior ML classifiers for stellar variability in the introduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive comments, which help strengthen the manuscript. We address each major comment below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The stated accuracies (94.26% Kepler, 88.22% TESS) are presented without any information on dataset splits, class balance, cross-validation strategy, preprocessing, or error analysis. These omissions directly affect evaluation of the central performance claims.

    Authors: We agree that the abstract, constrained by length, omits these methodological details. The full manuscript describes the data preparation in the Methods section, including an 80/10/10 train/validation/test split, handling of class imbalance via weighted loss, 5-fold cross-validation, preprocessing (normalization, gap-filling, and sigma-clipping), and error analysis (overall accuracy plus per-class precision/recall and confusion matrices). To improve accessibility without exceeding abstract limits, we will revise the abstract to include a concise statement on the validation strategy and dataset characteristics. revision: yes

  2. Referee: [Results/Deployment] Deployment and results sections: The application to ~2.8 million TESS light curves assumes cross-instrument generalization, yet no cross-mission hold-out experiments, label-source audits, or ablation studies addressing domain shift (differences in cadence, noise, and precision between Kepler and TESS) are described.

    Authors: The model was trained and validated on light curves from both missions, with the separate TESS accuracy of 88.22% providing direct evidence of performance on TESS data. We acknowledge that explicit cross-mission hold-out experiments and dedicated domain-shift ablations are not included. In the revised manuscript we will add a dedicated subsection in the Results discussing potential domain differences (cadence, noise properties) and include an ablation comparing Kepler-only versus combined training. Label sources are the standard Kepler and TESS variability catalogs, which we will explicitly audit and document. revision: yes

Circularity Check

0 steps flagged

No circularity: standard supervised ML on external labeled data

full rationale

The paper describes training and validating a Transformer+BiLSTM+CNN classifier on Kepler and TESS light-curve datasets with pre-existing labels, then deploying the trained model on a larger TESS sample. Reported accuracies (94.26% Kepler, 88.22% TESS) are empirical test-set performance metrics obtained after supervised optimization; they are not obtained by fitting a parameter to the output quantity itself, nor by any self-referential equation, ansatz smuggled via citation, or uniqueness theorem. No load-bearing self-citations, self-definitional steps, or renaming of known results appear in the abstract or described workflow. The derivation chain is the conventional ML pipeline (data ingestion → model training → evaluation → inference) and remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that supervised learning on labeled light curves can produce accurate classifications; the model weights are fitted during training but no new physical parameters or entities are introduced.

free parameters (1)
  • neural network weights and hyperparameters
    Learned during training on Kepler and TESS labeled data; specific values not provided in the abstract.
axioms (1)
  • domain assumption Labeled light curves from Kepler and TESS missions represent distinct and learnable classes of stellar variability
    This underpins the supervised training and reported accuracies.

pith-pipeline@v0.9.0 · 5542 in / 1491 out tokens · 69087 ms · 2026-05-10T17:07:02.565369+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

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

    astro-ph.SR 2026-04 conditional novelty 4.0

    Machine learning classification of TESS data for 6 million stars in the LOPS2 field identifies 28% as candidate variables after filtering out 72% instrumental signals, producing one of the largest automated variabilit...

Reference graph

Works this paper leans on

103 extracted references · 87 canonical work pages · cited by 1 Pith paper · 7 internal anchors

  1. [1]

    2021, Reviews of Modern Physics, 93, 015001, doi: 10.1103/RevModPhys.93.015001

    Aerts, C. 2021, Reviews of Modern Physics, 93, 015001, doi: 10.1103/RevModPhys.93.015001

    work page doi:10.1103/revmodphys.93.015001 2021

  2. [2]

    Aerts, C., Christensen-Dalsgaard, J., & Kurtz, D. W. 2010, Asteroseismology, doi: 10.1007/978-1-4020-5803-5

    work page doi:10.1007/978-1-4020-5803-5 2010

  3. [3]

    2024, Astronomy & Astrophysics, 692, R1 3 https://tess.mit.edu/qlp/

    Aerts, C., & Tkachenko, A. 2024, Astronomy & Astrophysics, 692, R1 3 https://tess.mit.edu/qlp/

    2024

  4. [4]

    S., & Hey, D

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

    2025

  5. [5]

    J., Kirk, J., Lam, K

    Armstrong, D. J., Kirk, J., Lam, K. W. F., et al. 2016, MNRAS, 456, 2260, doi: 10.1093/mnras/stv2836 Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33, doi: 10.1051/0004-6361/201322068 Astropy Collaboration, Price-Whelan, A. M., Sip˝ ocz, B. M., et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f 16

    work page doi:10.1093/mnras/stv2836 2016

  6. [6]

    2025, Ap&SS, 370, 72, doi: 10.1007/s10509-025-04460-5

    Audenaert, J. 2025, Ap&SS, 370, 72, doi: 10.1007/s10509-025-04460-5

    work page doi:10.1007/s10509-025-04460-5 2025

  7. [7]

    Villar, V. A. 2025, ICML 2025 Workshop on Foundation Models for Structured Data. https://arxiv.org/abs/2507.05333

    work page arXiv 2025

  8. [8]

    2022, A&A, 666, A76, doi: 10.1051/0004-6361/202243469

    Audenaert, J., & Tkachenko, A. 2022, A&A, 666, A76, doi: 10.1051/0004-6361/202243469

    work page doi:10.1051/0004-6361/202243469 2022

  9. [9]

    , keywords =

    Audenaert, J., Kuszlewicz, J. S., Handberg, R., et al. 2021, AJ, 162, 209, doi: 10.3847/1538-3881/ac166a

    work page doi:10.3847/1538-3881/ac166a 2021

  10. [10]

    Layer Normalization

    Ba, J. L., Kiros, J. R., & Hinton, G. E. 2016, Layer Normalization. https://arxiv.org/abs/1607.06450

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  11. [11]

    H., Bedding, T

    Barbara, N. H., Bedding, T. R., Fulcher, B. D., Murphy, S. J., & Van Reeth, T. 2022, MNRAS, 514, 2793, doi: 10.1093/mnras/stac1515

    work page doi:10.1093/mnras/stac1515 2022

  12. [12]

    2025, Multiband Embeddings of Light Curves

    Becker, I., Protopapas, P., Catelan, M., & Pichara, K. 2025, Multiband Embeddings of Light Curves. https://arxiv.org/abs/2501.12499

    work page arXiv 2025

  13. [13]

    , archivePrefix = "arXiv", eprint =

    Blomme, J., Sarro, L. M., O’Donovan, F. T., et al. 2011, MNRAS, 418, 96, doi: 10.1111/j.1365-2966.2011.19466.x

    work page doi:10.1111/j.1365-2966.2011.19466.x 2011

  14. [14]

    On the Opportunities and Risks of Foundation Models

    Bommasani, R., Hudson, D. A., Adeli, E., et al. 2021, arXiv preprint arXiv:2108.07258, arXiv:2108.07258, doi: 10.48550/arXiv.2108.07258

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2108.07258 2021

  15. [15]

    Kepler Planet-Detection Mission: Introduction and First Results.Science2010,327, 977

    Borucki, W. J., Koch, D., Basri, G., et al. 2010, Science, 327, 977, doi: 10.1126/science.1185402

    work page doi:10.1126/science.1185402 2010

  16. [16]

    2001, Machine Learning, 45, 5

    Breiman, L. 2001, Machine Learning, 45, 5

    2001

  17. [17]

    Y., Espinoza-Rojas, F., Copp´ ee, Q., & Hekker, S

    Choi, J. Y., Espinoza-Rojas, F., Copp´ ee, Q., & Hekker, S. 2025, A&A, 699, A180, doi: 10.1051/0004-6361/202555279

    work page doi:10.1051/0004-6361/202555279 2025

  18. [18]

    Gaia Data Release 3

    Clementini, G., Ripepi, V., Garofalo, A., et al. 2023, A&A, 674, A18, doi: 10.1051/0004-6361/202243964

    work page doi:10.1051/0004-6361/202243964 2023

  19. [19]

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

    Colman, I. L., Angus, R., David, T., et al. 2024, The Astronomical Journal, 167, 189

    2024

  20. [20]

    J., & Feng, F

    Cui, K., Armstrong, D. J., & Feng, F. 2024, ApJS, 274, 29, doi: 10.3847/1538-4365/ad62fd

    work page doi:10.3847/1538-4365/ad62fd 2024

  21. [21]

    Language Modeling with Gated Convolutional Networks

    Dauphin, Y. N., Fan, A., Auli, M., & Grangier, D. 2016, arXiv e-prints, arXiv:1612.08083, doi: 10.48550/arXiv.1612.08083 De Ridder, J., Ripepi, V., Aerts, C., et al. 2023, Astronomy & Astrophysics, 674, A36

    work page Pith review doi:10.48550/arxiv.1612.08083 2016

  22. [22]

    M., Aerts, C., et al

    Debosscher, J., Sarro, L. M., Aerts, C., et al. 2007, A&A, 475, 1159, doi: 10.1051/0004-6361:20077638

    work page doi:10.1051/0004-6361:20077638 2007

  23. [23]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. 2018, arXiv e-prints, arXiv:1810.04805, doi: 10.48550/arXiv.1810.04805

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1810.04805 2018

  24. [24]

    2026, A&A, 707, A170, doi: 10.1051/0004-6361/202554026 —

    Donoso-Oliva, C., Becker, I., Protopapas, P., et al. 2026, A&A, 707, A170, doi: 10.1051/0004-6361/202554026 —. 2023, A&A, 670, A54, doi: 10.1051/0004-6361/202243928

    work page doi:10.1051/0004-6361/202554026 2026

  25. [25]

    Eschen, Y. N. E., Bayliss, D., Wilson, T. G., et al. 2024, arXiv e-prints, arXiv:2409.13039, doi: 10.48550/arXiv.2409.13039

    work page doi:10.48550/arxiv.2409.13039 2024

  26. [26]

    2019, GitHub

    Falcon, W., & The PyTorch Lightning team. 2019, GitHub

    2019

  27. [27]

    2023, ApJS, 268, 4, doi: 10.3847/1538-4365/acdee5

    Fetherolf, T., Pepper, J., Simpson, E., et al. 2023, ApJS, 268, 4, doi: 10.3847/1538-4365/acdee5

    work page doi:10.3847/1538-4365/acdee5 2023

  28. [28]

    M., Tan, C

    Foumani, N. M., Tan, C. W., Webb, G. I., & Salehi, M. 2023, arXiv e-prints, arXiv:2305.16642, doi: 10.48550/arXiv.2305.16642

    work page doi:10.48550/arxiv.2305.16642 2023

  29. [29]

    Friedman, J. H. 2001, Annals of statistics, 1189

    2001

  30. [30]

    2025a, arXiv preprint arXiv:2512.09395

    Fritzewski, D., Kemp, A., Li, G., & Aerts, C. 2025a, arXiv preprint arXiv:2512.09395

    work page arXiv

  31. [31]

    Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning

    Gal, Y., & Ghahramani, Z. 2015, arXiv e-prints, arXiv:1506.02142, doi: 10.48550/arXiv.1506.02142

    work page Pith review doi:10.48550/arxiv.1506.02142 2015

  32. [32]

    Han, T., & Brandt, T. D. 2023, AJ, 165, 71, doi: 10.3847/1538-3881/acaaa7

    work page doi:10.3847/1538-3881/acaaa7 2023

  33. [33]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2

    work page doi:10.1038/s41586-020-2649-2 2020

  34. [34]

    B., Chaplin, W

    Hatt, E., Nielsen, M. B., Chaplin, W. J., et al. 2023, A&A, 669, A67, doi: 10.1051/0004-6361/202244579

    work page doi:10.1051/0004-6361/202244579 2023

  35. [35]

    2024, A&A, 688, A93, doi: 10.1051/0004-6361/202450489

    Hey, D., & Aerts, C. 2024, A&A, 688, A93, doi: 10.1051/0004-6361/202450489

    work page doi:10.1051/0004-6361/202450489 2024

  36. [36]

    Neural Computation 9(8), 1735–1780 (1997)

    Hochreiter, S., & Schmidhuber, J. 1997, Neural Computation, 9, 1735, doi: 10.1162/neco.1997.9.8.1735

    work page doi:10.1162/neco.1997.9.8.1735 1997

  37. [37]

    A., et al

    Hon, M., Stello, D., Garc´ ıa, R. A., et al. 2019, MNRAS, 485, 5616, doi: 10.1093/mnras/stz622

    work page doi:10.1093/mnras/stz622 2019

  38. [38]

    2018a, MNRAS, 476, 3233, doi: 10.1093/mnras/sty483

    Hon, M., Stello, D., & Yu, J. 2018a, MNRAS, 476, 3233, doi: 10.1093/mnras/sty483

    work page doi:10.1093/mnras/sty483

  39. [39]

    Hon, M., Stello, D., & Zinn, J. C. 2018b, ApJ, 859, 64, doi: 10.3847/1538-4357/aabfdb

    work page doi:10.3847/1538-4357/aabfdb

  40. [40]

    B., Sobeck, C., Haas, M., et al

    Howell, S. B., Sobeck, C., Haas, M., et al. 2014, PASP, 126, 398, doi: 10.1086/676406

    work page doi:10.1086/676406 2014

  41. [41]

    X., Vanderburg, A., P´ al, A., et al

    Huang, C. X., Vanderburg, A., P´ al, A., et al. 2020a, Research Notes of the American Astronomical Society, 4, 204, doi: 10.3847/2515-5172/abca2e —. 2020b, Research Notes of the American Astronomical Society, 4, 206, doi: 10.3847/2515-5172/abca2d

    work page doi:10.3847/2515-5172/abca2e

  42. [42]

    2025, arXiv e-prints, arXiv:2512.10002, doi: 10.48550/arXiv.2512.10002

    Huber, D. 2025, arXiv e-prints, arXiv:2512.10002, doi: 10.48550/arXiv.2512.10002

    work page doi:10.48550/arxiv.2512.10002 2025

  43. [43]

    2025, A&A, 701, A150, doi: 10.1051/0004-6361/202554025

    Huijse, P., De Ridder, J., Eyer, L., et al. 2025, A&A, 701, A150, doi: 10.1051/0004-6361/202554025

    work page doi:10.1051/0004-6361/202554025 2025

  44. [44]

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

    work page doi:10.1109/mcse.2007.55 2007

  45. [45]

    W., Tkachenko, A., Johnston, C., & Aerts, C

    IJspeert, L. W., Tkachenko, A., Johnston, C., & Aerts, C. 2024a, arXiv e-prints, arXiv:2409.20540, doi: 10.48550/arXiv.2409.20540

    work page doi:10.48550/arxiv.2409.20540

  46. [46]

    W., Tkachenko, A., Johnston, C., et al

    IJspeert, L. W., Tkachenko, A., Johnston, C., et al. 2021, A&A, 652, A120, doi: 10.1051/0004-6361/202141489 17 —. 2024b, A&A, 685, A62, doi: 10.1051/0004-6361/202349079

    work page doi:10.1051/0004-6361/202141489 2021

  47. [47]

    2015, in Proceedings of Machine Learning Research, Vol

    Ioffe, S., & Szegedy, C. 2015, in Proceedings of Machine Learning Research, Vol. 37, Proceedings of the 32nd International Conference on Machine Learning, ed. F. Bach & D. Blei (Lille, France: PMLR), 448–456. https://proceedings.mlr.press/v37/ioffe15.html

    2015

  48. [48]

    Jamal, S., & Bloom, J. S. 2020, ApJS, 250, 30, doi: 10.3847/1538-4365/aba8ff

    work page doi:10.3847/1538-4365/aba8ff 2020

  49. [49]

    2025, A&A, 694, A185, doi: 10.1051/0004-6361/202452811

    Jannsen, N., Tkachenko, A., Royer, P., et al. 2025, A&A, 694, A185, doi: 10.1051/0004-6361/202452811

    work page doi:10.1051/0004-6361/202452811 2025

  50. [50]

    Kemp, A., Vrancken, J., Mombarg, J. S. G., et al. 2025, A&A, 704, A280, doi: 10.1051/0004-6361/202557362

    work page doi:10.1051/0004-6361/202557362 2025

  51. [51]

    Kim, D.-W., & Bailer-Jones, C. A. L. 2016, A&A, 587, A18, doi: 10.1051/0004-6361/201527188

    work page doi:10.1051/0004-6361/201527188 2016

  52. [52]

    2025, A&A, 703, A240, doi: 10.1051/0004-6361/202556079

    Kliapets, M., Huijse, P., Tkachenko, A., et al. 2025, A&A, 703, A240, doi: 10.1051/0004-6361/202556079

    work page doi:10.1051/0004-6361/202556079 2025

  53. [53]

    G., Borucki, W

    Koch, D. G., Borucki, W. J., Basri, G., et al. 2010, ApJL, 713, L79, doi: 10.1088/2041-8205/713/2/L79

    work page doi:10.1088/2041-8205/713/2/l79 2010

  54. [54]

    Krizhevsky, A., Sutskever, I., & Hinton, G. E. 2012, in Advances in Neural Information Processing Systems, ed. F. Pereira, C. Burges, L. Bottou, & K. Weinberger, Vol. 25 (Curran Associates, Inc.). https://proceedings.neurips.cc/paper files/paper/2012/ file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf

    2012

  55. [55]

    2022, Research Notes of the American Astronomical Society, 6, 236, doi: 10.3847/2515-5172/aca158

    Kunimoto, M., Tey, E., Fong, W., et al. 2022, Research Notes of the American Astronomical Society, 6, 236, doi: 10.3847/2515-5172/aca158

    work page doi:10.3847/2515-5172/aca158 2022

  56. [56]

    2021, Research Notes of the American Astronomical Society, 5, 234, doi: 10.3847/2515-5172/ac2ef0

    Kunimoto, M., Huang, C., Tey, E., et al. 2021, Research Notes of the American Astronomical Society, 5, 234, doi: 10.3847/2515-5172/ac2ef0

    work page doi:10.3847/2515-5172/ac2ef0 2021

  57. [57]

    Kurtz, D. W. 2022, ARA&A, 60, 31, doi: 10.1146/annurev-astro-052920-094232

    work page doi:10.1146/annurev-astro-052920-094232 2022

  58. [58]

    Neural Computation , author =

    LeCun, Y., Boser, B., Denker, J. S., et al. 1989, Neural Computation, 1, 541, doi: 10.1162/neco.1989.1.4.541

    work page doi:10.1162/neco.1989.1.4.541 1989

  59. [59]

    1998 , month = nov, journal =

    LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. 1998, Proceedings of the IEEE, 86, 2278, doi: 10.1109/5.726791

    work page doi:10.1109/5.726791 1998

  60. [60]

    R., et al

    Li, G., Van Reeth, T., Bedding, T. R., et al. 2020, Monthly Notices of the Royal Astronomical Society, 491, 3586 Lightkurve Collaboration, Cardoso, J. V. d. M., Hedges, C., et al. 2018, Lightkurve: Kepler and TESS time series analysis in Python, Astrophysics Source Code Library, record ascl:1812.013

    2020

  61. [61]

    Lomb, N. R. 1976, Ap&SS, 39, 447, doi: 10.1007/BF00648343

    work page doi:10.1007/bf00648343 1976

  62. [62]

    Decoupled Weight Decay Regularization

    Loshchilov, I., & Hutter, F. 2017, arXiv e-prints, arXiv:1711.05101, doi: 10.48550/arXiv.1711.05101

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1711.05101 2017

  63. [63]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    McInnes, L., Healy, J., & Melville, J. 2018, arXiv e-prints, arXiv:1802.03426, doi: 10.48550/arXiv.1802.03426

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1802.03426 2018

  64. [64]

    2010, in Proceedings of the 9th Python in Science Conference, ed

    McKinney, W. 2010, in Proceedings of the 9th Python in Science Conference, ed. S. van der Walt & J. Millman, 51–56

    2010

  65. [65]

    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. https://openreview.net/forum?id=nebGk9bm3L

    2026

  66. [66]

    S., Aerts, C., Van Reeth, T., & Hey, D

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

    2024

  67. [67]

    2025, A&A, 703, A41, doi: 10.1051/0004-6361/202554289

    Moreno-Cartagena, D., Protopapas, P., Cabrera-Vives, G., et al. 2025, A&A, 703, A41, doi: 10.1051/0004-6361/202554289

    work page doi:10.1051/0004-6361/202554289 2025

  68. [68]

    S., Biswas, R., & Hloˇ zek, R

    Muthukrishna, D., Narayan, G., Mandel, K. S., Biswas, R., & Hloˇ zek, R. 2019, PASP, 131, 118002, doi: 10.1088/1538-3873/ab1609

    work page doi:10.1088/1538-3873/ab1609 2019

  69. [69]

    2025, A&A, 694, A313, doi: 10.1051/0004-6361/202452325

    Nascimbeni, V., Piotto, G., Cabrera, J., et al. 2025, A&A, 694, A313, doi: 10.1051/0004-6361/202452325

    work page doi:10.1051/0004-6361/202452325 2025

  70. [70]

    S., P´ erez, F., & van der Walt, S

    Naul, B., Bloom, J. S., P´ erez, F., & van der Walt, S. 2018, Nature Astronomy, 2, 151, doi: 10.1038/s41550-017-0321-z

    work page doi:10.1038/s41550-017-0321-z 2018

  71. [71]

    Davies, G. R. 2022, A&A, 663, A51, doi: 10.1051/0004-6361/202243064

    work page doi:10.1051/0004-6361/202243064 2022

  72. [72]

    K., Ishitani Silva, S., et al

    Olmschenk, G., Barry, R. K., Ishitani Silva, S., et al. 2024, AJ, 168, 83, doi: 10.3847/1538-3881/ad55f1

    work page doi:10.3847/1538-3881/ad55f1 2024

  73. [73]

    2024, MNRAS, 528, 5890, doi: 10.1093/mnras/stae068

    Pan, J.-S., Ting, Y.-S., & Yu, J. 2024, MNRAS, 528, 5890, doi: 10.1093/mnras/stae068

    work page doi:10.1093/mnras/stae068 2024

  74. [74]

    2024, MNRAS, 531, 4990, doi: 10.1093/mnras/stae1450

    Parker, L., Lanusse, F., Golkar, S., et al. 2024, MNRAS, 531, 4990, doi: 10.1093/mnras/stae1450

    work page doi:10.1093/mnras/stae1450 2024

  75. [75]

    2019, in Advances in Neural Information Processing Systems 32 (Curran

    Paszke, A., Gross, S., Massa, F., et al. 2019, in Advances in Neural Information Processing Systems 32 (Curran

    2019

  76. [76]

    2011, Journal of Machine Learning Research, 12, 2825

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825

    2011

  77. [77]

    2026, arXiv e-prints, arXiv:2603.22236

    Petitpas, G., Haviland, J., Han, T., et al. 2026, arXiv e-prints, arXiv:2603.22236. https://arxiv.org/abs/2603.22236

    work page arXiv 2026

  78. [78]

    Radford, A., Narasimhan, K., Salimans, T., & Sutskever, I. 2018

    2018

  79. [79]

    2025, A&A, 693, A268, doi: 10.1051/0004-6361/202452429

    Ranaivomanana, P., Uzundag, M., Johnston, C., et al. 2025, A&A, 693, A268, doi: 10.1051/0004-6361/202452429

    work page doi:10.1051/0004-6361/202452429 2025

  80. [80]

    2024, arXiv e-prints, arXiv:2410.16336, doi: 10.48550/arXiv.2410.16336

    Ranjbar, M., & Rahimzadeh, M. 2024, arXiv e-prints, arXiv:2410.16336, doi: 10.48550/arXiv.2410.16336

    work page doi:10.48550/arxiv.2410.16336 2024

Showing first 80 references.