arxiv: 2605.05324 · v1 · submitted 2026-05-06 · 🌌 astro-ph.IM · astro-ph.SR

Recognition: unknown

A useful representation of TESS light curves

Dovi Poznanski

Authors on Pith no claims yet

Pith reviewed 2026-05-08 16:08 UTC · model grok-4.3

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

keywords TESS light curvesself-organizing mapsquantile graphsprincipal component analysisvariability analysistime series representationexploratory data analysissignal-to-noise ratio

0 comments

The pith

A quantile-graph representation projected onto a self-organizing map organizes TESS light curves by amplitude, signal-to-noise ratio, timescale, and shape.

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

The paper develops a representation for TESS light curves that places them in a space where nearby points share similar variability properties. Each curve is encoded as a quantile graph, reduced in dimension by principal component analysis, and arranged on a self-organizing map. After testing roughly 1500 configurations with embedding diagnostics and a shape-cohesion metric, the authors select a compact quantile-graph version for its efficiency and stability. When applied to 1.5 million 2-minute cadence curves, the map groups sources mainly by variability amplitude, signal strength relative to noise, typical change timescale, and overall variation pattern. Repeated observations of the same stars remain in stable, contiguous regions, indicating the layout tracks lasting stellar traits instead of transient noise or systematics.

Core claim

What carries the argument

The self-organizing map built from principal components of quantile-graph encodings of the light curves. It places the reduced features on a two-dimensional grid so that proximity corresponds to similarity in variability amplitude, timescale, shape, and signal quality.

Load-bearing premise

The light-curve-shape cohesion metric and embedding diagnostics correctly identify a configuration in which map proximity reflects genuine similarity in stellar variability rather than artifacts of the selection process itself.

What would settle it

Finding that repeat observations of the same stars frequently land in distant or non-contiguous regions of the map would show that the representation fails to capture persistent properties.

Figures

Figures reproduced from arXiv: 2605.05324 by Dovi Poznanski.

**Figure 1.** Figure 1: — Comparison of representative model configurations across five evaluation metrics: shape cohesion, quantization error, topological error, trustworthiness, and continuity. Metrics are rescaled so that better performance lies toward the perimeter. Gray polygons show all ∼ 1500 evaluated configurations; colored curves highlight selected models discussed in the text. QG-based representations generally perfo… view at source ↗

**Figure 2.** Figure 2: — Node-level statistics for 3 SOMs during our hyperparameter search, the best QG (top row), the best ST (middle row), and one of the worst models for contrast (bottom row). Clearly, the more successful SOMs trace the periodic signals, and correlate with the period as well as the various measures of variability or SNR. topological error (Kiviluoto 1996), the trustworthiness, and the continuity (Venna and Ka… view at source ↗

**Figure 3.** Figure 3: — We map our main training set of 46K light curves to our fiducial SOM. Every node in the top panel is colored according to the median period of the training set in the node. We superpose in black the median-aligned, scaled, and stacked light curves in every node, with the scatter (±2 MAD) shown in gray. Below, we follow a progression of nodes from the bottom left toward the middle, showing the five light … view at source ↗

**Figure 4.** Figure 4: — Node-level statistics for the full 2-minute catalog. The SOM correlates strongly with various measures of variability amplitude and SNR, in close analogy to the training set ( view at source ↗

**Figure 5.** Figure 5: — Purity (estimated by eye) as a function of percentileranked distance from the node center, in bins of SNR. On average, Dbmu provides a useful indicator of confidence in node membership. 0 10 20 30 40 nsector 0 0.2 0.4 0.6 0.8 1 m e dia n ( n node nsector ) best limit - 1/nsector random limit measured ratio ' 1< view at source ↗

**Figure 6.** Figure 6: — We count how often repeat observations of the same star ended up in different nodes. If assignments were random the number of nodes would be close to the number of independent observations and the (blue) measurements would be near the dashed red line. Instead we are much closer to the red line, the optimal limit were all the sources are assumed to be persistent and noiseless and assignments are perfectl… view at source ↗

**Figure 8.** Figure 8: — Overview of the SOM in the web interface. Each hexagonal cell represents a node, colored here by a variability proxy. The map provides a global view of the organization of light-curves in the learned feature space, and an entry point to the catalog. Clicking on a node brings node-level information and a view of the light curves associated with it view at source ↗

**Figure 9.** Figure 9: — Example TIC-level view in the web interface. The page displays all the sectors where this star was observed and where they are mapped on the SOM. This allows to quickly identify objects that change state, or nodes that have similar sources view at source ↗

read the original abstract

We present a simple and interpretable representation of TESS light curves designed for large-scale exploratory analysis. Our goal is not to optimize classification performance, but to construct a computationally efficient mapping in which proximity reflects meaningful similarity, without using labels or explicit period information as inputs. We represent each light curve using either quantile graphs or scattering transforms, reduce dimensionality with principal component analysis, and project the resulting features onto a self-organizing map (SOM). We evaluate ~1500 model configurations using a combination of standard embedding diagnostics and a light-curve-shape-based cohesion metric, and select a compact quantile-graph-based model that balances interpretability, stability, and performance. Applying the model to ~1.5 million TESS 2-minute cadence light curves, we find that the map organizes sources primarily by variability amplitude, signal-to-noise ratio, characteristic timescale, and light-curve shape. Repeat observations of the same stars show that most sources occupy stable and contiguous regions of the map, indicating that the representation captures persistent properties rather than noise and systematics. We provide an interactive web interface at http://tess-l8.space that enables inspection of nodes, nearest neighbors, and individual sources across sectors. The resulting representation serves as a practical tool for exploration, anomaly detection, and dataset characterization, and illustrates how simple, deterministic encodings can yield useful structure in large astronomical time-series datasets.

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

A straightforward quantile-graph SOM embedding for 1.5M TESS curves that groups by amplitude, timescale and shape and holds on repeats, but the selection from 1500 configs leaves open whether the structure is robust or partly tuned to the internal metrics.

read the letter

The paper builds a label-free 2D map of TESS light curves by turning each into a quantile graph, running PCA, and laying the result on a self-organizing map. They test roughly 1500 combinations of those steps plus scattering transforms, pick one that scores well on standard embedding checks and a custom shape-cohesion measure, then apply it to 1.5 million 2-minute cadence curves. The map sorts sources mainly by variability amplitude, signal-to-noise, characteristic timescale, and light-curve shape, and repeat observations of the same stars land in stable, nearby regions. They also release an interactive viewer at tess-l8.space for browsing nodes and neighbors. That combination of scale, simplicity, and public access is the concrete deliverable. The stability on repeats is the strongest piece of evidence that the embedding is capturing persistent source properties rather than sector artifacts. The soft spot is the model-selection step. Because the final choice was made after seeing all 1500 runs and the cohesion metric itself rewards shape similarity, the reported organization by shape and the repeat stability could be inflated by post-hoc tuning. No pre-registered selection protocol or external labels are mentioned to break that loop. The work is aimed at astronomers who need a quick, interpretable way to scan or flag interesting objects in large TESS archives without training on labels. It is not positioned as a classifier or period finder, so it fits that exploratory niche. I would send it for peer review so referees can check the metric definitions and test whether the same organization appears under different selection rules.

Referee Report

2 major / 2 minor

Summary. The paper presents a representation for TESS light curves using quantile graphs (or scattering transforms), PCA dimensionality reduction, and a self-organizing map (SOM). After systematically evaluating ~1500 configurations with standard embedding diagnostics plus a custom light-curve-shape cohesion metric, the authors select a compact quantile-graph model. They apply it to ~1.5 million 2-minute TESS light curves and report that the resulting map organizes sources primarily by variability amplitude, signal-to-noise ratio, characteristic timescale, and light-curve shape. Repeat observations of the same stars are shown to occupy stable, contiguous regions, suggesting the representation captures persistent source properties. An interactive web interface is provided for exploration.

Significance. If the central claims hold after addressing selection concerns, the work offers a practical, computationally efficient, and interpretable tool for large-scale exploratory analysis of TESS time-series data. Strengths include the scale of application (1.5M light curves), the public interactive interface, emphasis on deterministic encodings without labels or periods, and potential utility for anomaly detection and dataset characterization. This aligns with needs in astro-ph.IM for methods that reveal structure in high-volume survey data.

major comments (2)

[Methods (model evaluation and selection)] Methods section describing model selection: The final configuration is chosen after evaluating ~1500 variants using a combination of standard diagnostics and a light-curve-shape-based cohesion metric. Because this metric is defined in terms of shape similarity (one of the four claimed organizing axes) and selection occurs after inspecting all results, the observed organization by amplitude, SNR, timescale, and shape on the 1.5M sources, as well as the repeat-observation stability, could be inflated by post-hoc optimization rather than reflecting an intrinsic property of the encoding. The manuscript should specify whether selection criteria were pre-registered, whether held-out data or external labels were used during choice, and report quantitative scores (e.g., cohesion values, embedding quality metrics) for the selected model versus representative alternatives.
[Results (application and stability analysis)] Results section on application to 1.5M light curves: The claim that proximity in the map reflects meaningful similarity rests on the selected SOM being a faithful embedding. Without explicit details on data exclusion rules, sector coverage, or how sources with multiple observations were chosen for the stability test, it is unclear whether the reported organization and contiguity are robust to reasonable variations in the input sample.

minor comments (2)

[Abstract and Results] The abstract states that the representation 'organizes sources primarily by' four properties, but the main text should include quantitative support (e.g., correlation coefficients or variance explained by each axis) rather than qualitative description alone.
[Discussion or Conclusions] The interactive interface at http://tess-l8.space is a valuable contribution; a brief description of its features (node inspection, nearest-neighbor search, sector navigation) should be added to the main text or a dedicated subsection for readers who do not immediately access the site.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive review and positive assessment of the manuscript's potential utility. We address each major comment below and will revise the manuscript accordingly to improve clarity and address concerns about selection and robustness.

read point-by-point responses

Referee: [Methods (model evaluation and selection)] The final configuration is chosen after evaluating ~1500 variants using a combination of standard diagnostics and a light-curve-shape-based cohesion metric. Because this metric is defined in terms of shape similarity and selection occurs after inspecting all results, the observed organization could be inflated by post-hoc optimization. The manuscript should specify whether selection criteria were pre-registered, whether held-out data or external labels were used, and report quantitative scores for the selected model versus alternatives.

Authors: We acknowledge that the cohesion metric, being based on shape similarity, could introduce some dependence when assessing the shape axis. However, the other organizing axes (amplitude, SNR, and timescale) were evaluated using independent standard embedding diagnostics such as trustworthiness, continuity, and neighborhood preservation, none of which rely on the cohesion metric. Selection criteria were not pre-registered, as this practice is not standard for unsupervised exploratory methods in astronomy; no external labels were used at any stage. We will add a dedicated subsection in the Methods section that reports the full set of quantitative scores (cohesion, trustworthiness, etc.) for the selected model alongside several representative alternatives (e.g., the next-best models by each metric). This will allow readers to assess the sensitivity of the final choice. revision: yes
Referee: [Results (application and stability analysis)] Without explicit details on data exclusion rules, sector coverage, or how sources with multiple observations were chosen for the stability test, it is unclear whether the reported organization and contiguity are robust to reasonable variations in the input sample.

Authors: We will expand the Results section (and add a brief methods subsection on data preparation) to specify the exact exclusion rules applied to the 1.5 million light curves, including quality-flag thresholds, minimum cadence coverage, and any sector-specific filters. The sample comprises all publicly available 2-minute TESS sectors at the time of analysis. For the stability test, we included every source with observations in two or more sectors and used the chronologically first two observations per source; we will also report a supplementary check using randomly selected pairs to confirm that contiguity is not sensitive to this choice. These additions will make the robustness explicit. revision: yes

Circularity Check

0 steps flagged

No significant circularity; representation constructed bottom-up from data features without definitional reduction.

full rationale

The paper builds the representation via quantile graphs or scattering transforms, followed by PCA dimensionality reduction and SOM projection. Model selection among ~1500 configurations relies on standard embedding diagnostics plus a custom cohesion metric, after which the map is applied to 1.5M light curves and observed to organize by amplitude, SNR, timescale, and shape, with stability confirmed on repeat observations. No equations are presented that define the output organization in terms of the selection metric itself, no self-citations are load-bearing, and no ansatz or uniqueness theorem is invoked to force the result. The derivation remains self-contained and data-driven rather than tautological.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The approach rests on standard machine-learning primitives whose hyperparameters are tuned via the reported configuration search. No new physical entities are postulated.

free parameters (2)

PCA dimensionality
Selected during the search over ~1500 configurations to balance compactness and performance.
SOM grid size and training parameters
Chosen via embedding diagnostics and the cohesion metric.

axioms (2)

domain assumption Quantile graphs capture sufficient information about light-curve shape and amplitude for similarity assessment.
Used as the primary feature representation without explicit period or label inputs.
standard math Self-organizing maps arrange inputs so that proximity corresponds to feature-space similarity.
Core assumption underlying the final projection step.

pith-pipeline@v0.9.0 · 5534 in / 1515 out tokens · 69692 ms · 2026-05-08T16:08:58.138758+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

60 extracted references · 35 canonical work pages · 1 internal anchor

[1]

Private Communications

Me, Moi. Private Communications. OJA. doi:000
[2]

arXiv e-prints , keywords =

Microseismic Noise Mitigation with Machine Learning for Advanced LIGO. arXiv e-prints , keywords =. doi:10.48550/arXiv.2511.19682 , archivePrefix =. 2511.19682 , primaryClass =

work page doi:10.48550/arxiv.2511.19682
[3]

arXiv e-prints , keywords =

Exploration of groups and outliers in Gaia RVS stellar spectra with metric learning. arXiv e-prints , keywords =. doi:10.48550/arXiv.2508.00071 , archivePrefix =. 2508.00071 , primaryClass =

work page doi:10.48550/arxiv.2508.00071
[4]

, keywords =

Using Anomaly Detection to Search for Technosignatures in Breakthrough Listen Observations. , keywords =. doi:10.3847/1538-3881/add52b , archivePrefix =. 2505.03927 , primaryClass =

work page doi:10.3847/1538-3881/add52b
[5]

, keywords =

A semisupervised machine learning search for never-seen gravitational-wave sources. , keywords =. doi:10.1093/mnras/staa3550 , archivePrefix =. 2010.11949 , primaryClass =

work page doi:10.1093/mnras/staa3550 2010
[6]

Astronomy and Computing , keywords =

Effectively using unsupervised machine learning in next generation astronomical surveys. Astronomy and Computing , keywords =. doi:10.1016/j.ascom.2020.100437 , archivePrefix =. 1911.06823 , primaryClass =

work page doi:10.1016/j.ascom.2020.100437 2020
[7]

, keywords =

Redshifted broad absorption line quasars found via machine-learned spectral similarity. , keywords =. doi:10.1093/mnras/sty2127 , archivePrefix =. 1805.09829 , primaryClass =

work page doi:10.1093/mnras/sty2127
[8]

, keywords =

Detecting outliers and learning complex structures with large spectroscopic surveys - a case study with APOGEE stars. , keywords =. doi:10.1093/mnras/sty348 , archivePrefix =. 1711.00022 , primaryClass =

work page doi:10.1093/mnras/sty348
[9]

, keywords =

The weirdest SDSS galaxies: results from an outlier detection algorithm. , keywords =. doi:10.1093/mnras/stw3021 , archivePrefix =. 1611.07526 , primaryClass =

work page doi:10.1093/mnras/stw3021
[10]

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

Variability Catalog of Stars Observed during the TESS Prime Mission. , keywords =. doi:10.3847/1538-4365/acdee5 , archivePrefix =. 2208.11721 , primaryClass =

work page doi:10.3847/1538-4365/acdee5
[11]

arXiv e-prints , keywords =

Multilayer Quantile Graph for Multivariate Time Series Analysis and Dimensionality Reduction. arXiv e-prints , keywords =. doi:10.48550/arXiv.2311.11849 , archivePrefix =. 2311.11849 , primaryClass =

work page doi:10.48550/arxiv.2311.11849
[12]

arXiv e-prints , keywords =

Time Series Analysis via Network Science: Concepts and Algorithms. arXiv e-prints , keywords =. doi:10.48550/arXiv.2110.09887 , archivePrefix =. 2110.09887 , primaryClass =

work page doi:10.48550/arxiv.2110.09887
[13]

arXiv e-prints , keywords =

Group Invariant Scattering. arXiv e-prints , keywords =. doi:10.48550/arXiv.1101.2286 , archivePrefix =. 1101.2286 , primaryClass =

work page doi:10.48550/arxiv.1101.2286
[14]

1995 , doi =

Kohonen, Teuvo , title =. 1995 , doi =

1995
[15]

Campanharo, Andriana S. L. O. and Sirer, M. Irmak and Malmgren, R. Dean and Ramos, Fernando M. and Amaral, Luís A. Nunes , title =. PLoS ONE , volume =. 2011 , doi =

2011
[16]

Proceedings of the IEEE International Conference on Neural Networks (ICNN'96) , volume =

Kiviluoto, Kimmo , title =. Proceedings of the IEEE International Conference on Neural Networks (ICNN'96) , volume =. 1996 , doi =

1996
[17]

Advances in Self-Organizing Maps , editor =

Ultsch, Alfred and Siemon, Hans Peter , title =. Advances in Self-Organizing Maps , editor =. 2009 , doi =

2009
[18]

Proceedings of the Workshop on Self-Organizing Maps (WSOM'05) , pages =

Venna, Jarkko and Kaski, Samuel , title =. Proceedings of the Workshop on Self-Organizing Maps (WSOM'05) , pages =. 2005 , url =

2005
[19]

Proceedings of the 11th International Conference on Artificial Intelligence and Statistics (AISTATS) , pages =

Venna, Jarkko and Kaski, Samuel , title =. Proceedings of the 11th International Conference on Artificial Intelligence and Statistics (AISTATS) , pages =. 2007 , url =

2007
[20]

Jenkins, J. M. and Twicken, J. D. and McCauliff, S. and Campbell, J. and Sanderfer, D. and Lung, D. and others , title =. Proceedings of the SPIE , volume =. 2016 , doi =

2016
[21]

Smith, J. C. and Stumpe, M. C. and Van Cleve, J. E. and Jenkins, J. M. and Barclay, T. and Fanelli, M. N. and others , title =. Publications of the Astronomical Society of the Pacific , volume =. 2012 , doi =

2012
[22]

Stumpe, M. C. and Smith, J. C. and Van Cleve, J. E. and Twicken, J. D. and Barclay, T. and Fanelli, M. N. and others , title =. Publications of the Astronomical Society of the Pacific , volume =. 2012 , doi =

2012
[23]

Stumpe, M. C. and Smith, J. C. and Catanzarite, J. H. and Van Cleve, J. E. and Jenkins, J. M. and Twicken, J. D. and others , title =. Publications of the Astronomical Society of the Pacific , volume =. 2014 , doi =

2014
[24]

2000, A&AS, 143, 9, doi: 10.1051/aas:2000332

The SIMBAD astronomical database. The CDS reference database for astronomical objects. , keywords =. doi:10.1051/aas:2000332 , archivePrefix =. astro-ph/0002110 , primaryClass =

work page Pith review doi:10.1051/aas:2000332
[25]

and Winn, Joshua N

Ricker, George R. and Winn, Joshua N. and Vanderspek, Roland and others , title =. Journal of Astronomical Telescopes, Instruments, and Systems , year =
[26]

and Kulkarni, Shrinivas R

Bellm, Eric C. and Kulkarni, Shrinivas R. and Graham, Matthew J. and others , title =. Publications of the Astronomical Society of the Pacific , year =
[27]

and Starr, Dan L

Richards, Joseph W. and Starr, Dan L. and Butler, Nathaniel R. and others , title =. The Astrophysical Journal , year =
[28]

The Astrophysical Journal Supplement Series , year =

Fetherolf, Tara and Pepper, Joshua and Simpson, Emilie and others , title =. The Astrophysical Journal Supplement Series , year =
[29]

The Astrophysical Journal Supplement Series , year =

Gao, Xinyi and Chen, Xiaodian and Wang, Shu and Liu, Jifeng , title =. The Astrophysical Journal Supplement Series , year =
[30]

Naul, Brett and Bloom, Joshua S. and P. A Recurrent Neural Network for Classification of Unevenly Sampled Variable Stars , journal =. 2018 , volume =

2018
[31]

2019, arXiv preprint arXiv:1904.07248

Baron, Dalya , title =. arXiv e-prints , year =. 1904.07248 , archivePrefix =

work page arXiv 1904
[32]

Astrophysics and Space Science , year =

Audenaert, Jeroen , title =. Astrophysics and Space Science , year =
[33]

arXiv e-prints , year =

Junell, Alexandra and Sasli, Argyro and Fontinele Nunes, Felipe and others , title =. arXiv e-prints , year =. 2507.16088 , archivePrefix =

work page arXiv
[34]

and Prince, Thomas A

Finkbeiner, Douglas P. and Prince, Thomas A. and Whitebook, Samuel E. , title =. arXiv e-prints , year =. 2502.00243 , archivePrefix =

work page arXiv
[35]

and Armstrong, David J

Battley, Matthew P. and Armstrong, David J. and Pollacco, Don , title =. Monthly Notices of the Royal Astronomical Society , year =
[36]

and Poznanski, D

Pardo, S. and Poznanski, D. and Croft, S. and Siemion, A. P. V. and Lebofsky, M. , title =. The Astronomical Journal , year =
[37]

and Poznanski, D

Marianer, T. and Poznanski, D. and Prochaska, J. X. , title =. Monthly Notices of the Royal Astronomical Society , year =
[38]

and Rotman, M

Reis, I. and Rotman, M. and Poznanski, D. and Prochaska, J. X. and Wolf, L. , title =. Astronomy and Computing , year =
[39]

and Poznanski, D

Reis, I. and Poznanski, D. and Baron, D. and Zasowski, G. and Shahaf, S. , title =. Monthly Notices of the Royal Astronomical Society , year =
[40]

and Poznanski, D

Baron, D. and Poznanski, D. , title =. Monthly Notices of the Royal Astronomical Society , year =
[41]

Ivezi ´c, S

Ivezi. LSST: From Science Drivers to Reference Design and Anticipated Data Products , journal =. 2019 , volume =. doi:10.3847/1538-4357/ab042c , archivePrefix =

work page doi:10.3847/1538-4357/ab042c 2019
[42]

Ashley and Berger, Edo , title =

Zabriskie, Connor and Villar, V. Ashley and Berger, Edo , title =. The Astrophysical Journal , year =. doi:10.3847/1538-4357/acab67 , adsurl =

work page doi:10.3847/1538-4357/acab67
[43]

Ashley and Cranmer, Kyle and Berger, Edo , title =

Villar, V. Ashley and Cranmer, Kyle and Berger, Edo , title =. The Astrophysical Journal , year =. doi:10.3847/1538-4357/abfcd9 , adsurl =

work page doi:10.3847/1538-4357/abfcd9
[44]

Soraisam, M. D. and others , title =. Monthly Notices of the Royal Astronomical Society , year =. doi:10.1093/mnras/staa3023 , adsurl =

work page doi:10.1093/mnras/staa3023
[45]

Brett, D. R. and West, R. G. and Wheatley, P. J. , title =. Monthly Notices of the Royal Astronomical Society , year =. doi:10.1093/mnras/353.2.369 , url =

work page doi:10.1093/mnras/353.2.369
[46]

Armstrong, D. J. and Kirk, J. and Lam, K. W. F. and McCormac, J. and Osborn, H. P. and Spake, J. and Walker, S. and Brown, D. J. A. and Kristiansen, M. H. and Pollacco, D. and West, R. and Wheatley, P. J. , title =. Monthly Notices of the Royal Astronomical Society , year =. doi:10.1093/mnras/stv2836 , eprint =

work page doi:10.1093/mnras/stv2836
[47]

M., Aerts, C., et al

Debosscher, J. and Sarro, L. M. and Aerts, C. and L. Automated supervised classification of variable stars. I. Methodology , journal =. 2007 , volume =. doi:10.1051/0004-6361:20077638 , eprint =

work page doi:10.1051/0004-6361:20077638 2007
[48]

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

Audenaert, Jeroen and Tkachenko, Andrew , title =. Astronomy & Astrophysics , year =. doi:10.1051/0004-6361/202243469 , eprint =

work page doi:10.1051/0004-6361/202243469
[49]

Crake, D. A. and Mart. Linking Anomalous Behaviour with Stellar Properties: An Unsupervised Exploration of TESS Light Curves , journal =. 2023 , eprint =. doi:10.48550/arXiv.2301.10264 , url =

work page doi:10.48550/arxiv.2301.10264 2023
[50]

Scalable End-to-end Recurrent Neural Network for Variable star classification , journal =

Becker, Ignacio and Pichara, Karim and Catelan, M. Scalable End-to-end Recurrent Neural Network for Variable star classification , journal =. 2020 , volume =. doi:10.1093/mnras/staa350 , eprint =

work page doi:10.1093/mnras/staa350 2020
[51]

H., Bedding, T

Barbara, Nicholas H. and Bedding, Timothy R. and Fulcher, Ben D. and Murphy, Simon J. and Van Reeth, Timothy , title =. Monthly Notices of the Royal Astronomical Society , year =. doi:10.1093/mnras/stac1515 , eprint =

work page doi:10.1093/mnras/stac1515
[52]

ASTROMER: A transformer-based embedding for the representation of light curves , journal =

Donoso-Oliva, Crist. ASTROMER: A transformer-based embedding for the representation of light curves , journal =. 2023 , volume =. doi:10.1051/0004-6361/202243928 , eprint =

work page doi:10.1051/0004-6361/202243928 2023
[53]

2025, arXiv e-prints, arXiv:2502.02717, doi: 10.48550/arXiv.2502.02717 —

Donoso-Oliva, Crist. Astromer 2 , journal =. 2025 , eprint =. doi:10.48550/arXiv.2502.02717 , url =

work page doi:10.48550/arxiv.2502.02717 2025
[54]

, title =

Rizhko, Maria and Bloom, Joshua S. , title =. arXiv e-prints , year =. doi:10.48550/arXiv.2411.08842 , url =. 2411.08842 , archivePrefix =

work page doi:10.48550/arxiv.2411.08842
[55]

2020, MNRAS, 498, 5972, doi: 10.1093/mnras/staa2745

Nardiello, D , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2020 , month =. doi:10.1093/mnras/staa2745 , url =

work page doi:10.1093/mnras/staa2745 2020
[56]

Mahabal, Ashish and Djorgovski, S. G. and Drake, A. J. and Donalek, C. and Graham, M. J. and Williams, R. , title =. Bulletin of the Astronomical Society of India , year =. 1111.0976 , archivePrefix =

work page arXiv
[57]

Mahabal, Ashish and Djorgovski, S. G. and Drake, A. J. and Donalek, C. and Graham, M. J. and Williams, R. , title =. Astronomische Nachrichten , year =. 1201.0225 , archivePrefix =

work page arXiv
[58]

1877 , publisher =

Tolstoy, Leo , title =. 1877 , publisher =
[59]

, keywords =

The Revised TESS Input Catalog and Candidate Target List. , keywords =. doi:10.3847/1538-3881/ab3467 , archivePrefix =. 1905.10694 , primaryClass =

work page doi:10.3847/1538-3881/ab3467 1905
[60]

The TESS All-Sky Rotation Survey: Periods for 1,046,317 Stars Within 500 pc

The TESS All-Sky Rotation Survey: Periods for 944,056 Stars Within 500 pc. arXiv e-prints , keywords =. doi:10.48550/arXiv.2603.05586 , archivePrefix =. 2603.05586 , primaryClass =

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2603.05586