Pith. sign in

REVIEW 3 major objections 5 minor 40 references

Using anomaly detection to search for technosignatures in Breakthrough Listen observations

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that a simulation-calibrated anomaly-detection pipeline cuts a 10^11-spectrogram SETI search to a few thousand candidates and that none of them survive human scrutiny.

desk verdict A solid, honest null-result SETI paper with a genuinely new triage pipeline; the improvement claim is plausible but lacks statistical rigor, and the null is conditional on the simulated signal family. read the letter →

arxiv 2505.03927 v1 pith:X23V34O2 submitted 2025-05-06 astro-ph.IM

Snir Pardo , Dovi Poznanski , Steve Croft , Andrew P. V. Siemion , Matthew Lebofsky This is my paper · ORCID
classification astro-ph.IM
keywords technosignaturesanomalydetectionradiofrequencyinterferenceBreakthroughListenUMAPcandidatetriageSETImachinelearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that a deliberately simple, simulation-calibrated anomaly-detection pipeline can reduce the search for narrow-band radio technosignatures from roughly $\sim 10^{11}$ 220-Hz spectrograms to a few thousand that need human attention. It validates the two ranking scores it introduces by showing that the top one percent of candidates contain about 7 percent (frequency score), 14 percent (similarity score), and 22 percent (combined) genuinely promising top candidates, against under 3 percent for random selection. Applying the full pipeline to more than 800 nearby stars observed by Breakthrough Listen, the authors visually inspected about 20,000 spectrograms and report that no candidate survived scrutiny. If correct, the paper establishes that a minimal-assumption machine-learning triage is a practical and markedly better-than-random front end for SETI candidate vetting, even though it finds no signal.

What carries the argument

The load-bearing object is the simulated on-target-only region in a UMAP plane. For each cadence of six 80-by-16-pixel spectrograms, the pipeline computes pairwise cross-correlations, extracts 15 summary statistics, and projects them to two UMAP coordinates; kernel density estimation trained on 640,000 Setigen simulations defines a probability that a cadence belongs to the 'only on-target' category, and a threshold of 0.0618 selects candidates. Two additional scores then rank survivors: a frequency score from bagged Gaussian mixture models fitted to candidate density per band, and a similarity score comparing intra-on-target to on-off distances in a second UMAP embedding. The mechanism's role is to convert the expensive search into a small, rank-ordered list without assuming a specific drift rate or signal morphology.

What would settle it

Insert narrow-band synthetic signals with time-frequency morphologies absent from Setigen's profile library into real off-source cadences and run the pipeline; if the top-percentile 'top-candidate' fraction drops to the random baseline, the simulation-defined accepted region is too narrow and the null result is not informative for those signal classes.

Watch

Extended reading notes

Core claim

The central claim is that the two-stage filter pipeline materially improves candidate selection for human vetting, and that on the Breakthrough Listen sample it yields no surviving technosignature. Each six-pointing observing cadence is compressed to 15 cross-correlation statistics, mapped with UMAP into a two-dimensional plane, and accepted only if it falls in the region where simulated on-target-only signals cluster, which cuts the candidate pool by a factor of about $10^5$. The survivors are then ranked twice: a frequency score prefers candidates in radio-quiet parts of the band, and a similarity score prefers cadences whose on-target observations resemble each other in UMAP space while staying distant from off-target observations. The paper measures top-candidate fractions of about 7, 14, and 22 percent for the frequency, similarity, and combined rankings versus below 3 percent for random selection, and reports that roughly 20,000 inspected candidates, including about 2,000 second-stage inspections, were all ruled out as radio-frequency interference or instrumental noise.

Load-bearing premise

The pipeline's accepted region is learned from simulated signals generated with the Setigen package against one S-band background, so the whole search assumes that real extraterrestrial narrow-band signals resemble that simulation family closely enough to land in the same UMAP cluster.

Editorial extensions

If this is right

  • A cadence-only statistic set, not a signal-shape template, can cut SETI candidate lists by five orders of magnitude before human vetting.
  • Ranking by both quietness in frequency and temporal self-similarity outperforms either score alone, so future searches should combine the two axes.
  • Known promising candidates from a comparable deep-learning search mostly pass through the new filter and receive high scores, implying the two approaches select a consistent population.
  • The two candidates lost by the cross-correlation filter had the highest drift rates and low signal-to-noise ratios, marking a boundary in drift sensitivity.
  • No candidate survived second-stage inspection, so if the simulation family is representative, the Breakthrough Listen sample contains no narrow-band on-target-only signal of the modeled kinds.

Reading between the lines

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

  • Because the frequency score discards the noisiest 5 percent of star-band combinations, a real strong signal in an RFI-dense band would be deliberately sacrificed; a pipeline variant that only down-ranks rather than discards those combinations could be tested against this sample.
  • The similarity score's definition of 'persistent' is scale-bound at 220 Hz by 100 s, so a multi-scale version that zooms out automatically would likely catch drifting signals the current first filter misses, as the paper's two lost high-drift candidates hint.
  • The validation strategy is relative (versus random selection), so the 22-percent top-candidate rate is not an absolute detection probability; comparing against a conventional TurboSETI-ranked sample would show whether the gain is unique to this pipeline.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This manuscript presents LE2.0, an anomaly-detection pipeline for technosignature searches in Breakthrough Listen radio observations. The pipeline reduces roughly 10^11 spectrograms to a few thousand human-vetted candidates through three stages: a cross-correlation/UMAP/KDE filter that selects cadences with signals appearing only in on-target pointings, with the accepted region calibrated by Setigen simulations; a frequency score based on GMM density estimation that up-weights candidates in RFI-quiet spectral regions; and a similarity score based on UMAP distances that rewards on-target spectrograms that cluster together and away from off-target pointings. The authors report that the frequency score, similarity score, and combined score produce top-candidate fractions of about 7%, 14%, and 22% during human vetting, versus less than 3% for random selection, and that no candidate survived detailed scrutiny. They also compare their candidate list with the deep-learning search of Ma et al. (2023).

Significance. If the validation claims survive scrutiny, the paper offers a useful triage method for SETI data and a substantial null result from a large Breakthrough Listen sample. Its strengths are the scale of the analysis (roughly 10^11 spectrograms), the explicit description of the human vetting stages, the use of simulations to calibrate the UMAP accepted region, and the direct comparison with an independent machine-learning search. The main caveats are that the quantitative validation is carried out inside the same simulation family used to define the accepted region, and that the headline improvement over random selection is reported without uncertainty estimates or significance tests. These issues are fixable with additional analysis rather than fundamental flaws.

major comments (3)
  1. [Section 4] The central claim that the filters "significantly improve" candidate quality is not quantified. State the exact number of candidates in each of the four sub-samples and provide binomial confidence intervals or a two-proportion test comparing 7%, 14%, and 22% against the control value of less than 3%. Without these, the reader cannot tell whether the differences are within sampling noise, and the word "significant" is not supported. Also specify whether the same candidates can appear in more than one sub-sample, since the three scored samples may not be independent.
  2. [Sections 3.1 and 4] The accepted "only on-target" region is defined and validated with Setigen simulations that use amplitudes uniform in [0, 4 times the noise maximum], all available Setigen time/frequency profiles, and signals present in every A panel of the cadence. Because the validation in Section 4 uses the same simulation family, the reported top-candidate fractions are in-family recovery rates. Add out-of-family injections into real data—for example signals in only one or two A panels, amplitudes below the simulated range, higher drift rates, and morphologies not available in Setigen—and measure recovery rates after each filter. This is needed to support both the triage-quality claim and the null conclusion. The self-identified caveat in Section 4 that "any simulation will be obviously biased by the populations we choose to simulate" is appropriate but does not replace such tests.
  3. [Section 3.1] The probability threshold of 0.0618 for the "only on-target" category is described as somewhat arbitrary and highly dependent on the KDE bandwidth, and the statements that the UMAP hyper-parameter choices do not affect the results are not demonstrated. Provide a sensitivity analysis showing how the threshold and hyper-parameters change the number of candidates surviving the first filter, the 7%/14%/22% validation fractions, and the final null result. Alternatively, define the threshold from a pre-specified acceptable contamination rate rather than by eye from Figure 4.
minor comments (5)
  1. [Section 2] The scikit-learn implementation is cited as "SCIKIT-LEARN (de Boor 2001)"; the correct citation for the software is Pedregosa et al. (2011), with de Boor (2001) cited only for the underlying B-spline mathematics.
  2. [Section 2] The relation between "about 10^7 of these per star-band combination" and "~10^11 spectrograms in total" is not derived. State the number of star-band combinations and clarify whether the counts refer to 80x16 spectrogram images or to full six-panel cadences.
  3. [Section 4] "These number were selected" should read "These numbers were selected," and the term "top-candidate" should be defined consistently as a candidate that passes the first vetting stage; the current text uses it before that definition is fully established.
  4. [Section 3.3] Equation (1) assumes a complete six-panel cadence. State what is done if a pointing is missing or flagged as unusable, since the formulas and the UMAP embedding may otherwise receive incomplete inputs.
  5. [Figure 4] The histogram on the y-axis is labeled "Counts" but the sample size, bin width, and the number of real targets used to build it are not given; these details would make the threshold selection more reproducible.

Circularity Check

1 steps flagged · score 6.0 of 10

Similarity-score validation is self-definitional: the score encodes the same on/off pattern used to define 'top candidates', so part of the claimed improvement reduces to the construct itself; the frequency-score result and the null result remain independent.

  1. self definitional [Section 3.3, Equation (1), and Section 4 (human vetting definition)]
    "We expect a real ET signal to be similar to itself in all the 'A' observations, and be dissimilar from the off-target observations. ... A good candidate will have all the 'on-target' observations similar to each other, and therefore they will be close to each other in the UMAP space, relative to the distances to the 'off-target'. This ratio is the similarity score for a candidate"

    The validation label 'top-candidate' is defined in Section 4 by exactly the same on/off criterion that the similarity score is built to measure: cadences are discarded when they show 'signals in more than just the on-target panels', and surviving cadences are 'top-candidates'. Equation (1) is a smooth surrogate for that same only-on-target pattern. Therefore the reported 14% top-candidate fraction for the similarity-only sample is an internal consistency check — how well the score recovers its own defining pattern — not an independent validation that the score improves candidate quality. The frequency score (7%) is an independent axis based on RFI density, so the circularity is partial; the combined 22% score inherits the self-definitional component.

full rationale

The paper's central validation claim is that its filters improve the quality of candidates sent to human vetting relative to random selection, with quality measured by the fraction of 'top-candidates'. The similarity score is explicitly constructed from the same on-target/off-target separation that defines a top-candidate, so the similarity-score component of the validation is partly circular: it demonstrates that the score agrees with the human application of the same heuristic, not that the heuristic has external predictive value. The frequency-score result is independent, since it ranks by quietness of the frequency window rather than by the on/off pattern, and the paper's null result is an honest non-detection within the pipeline's accepted region. I weighed the paper's own limitation note — 'Any simulation will be obviously biased by the populations we choose to simulate' — as a coverage caveat about the Setigen-based first filter, not as a circular step, because no equation or fitted parameter reduces the null result to the simulation input. There is no load-bearing self-citation chain: citations to prior work by authors are contextual examples, and the Setigen package is public tooling, not an imported uniqueness theorem. Overall, one prediction metric reduces by construction to its target definition, giving partial circularity rather than complete equivalence, hence a score of 6.

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

The central claims rest on the cadence-based definition of a candidate, a hand-tuned simulation calibration, and a human-labeled vetting benchmark. No external ground truth for ET signals exists, so the paper pays for most of its discriminatory power through domain choices rather than first-principles derivations. The frequency score is the most independent component; the similarity score and the validation labels share the same on/off criterion.

free parameters (6)
  • Only on-target probability threshold = 0.0618
    Chosen by eye from the distribution of probabilities for 20 real targets; the authors call it somewhat arbitrary and note it depends on the KDE bandwidth.
  • UMAP hyperparameters = n_neighbors=25, min_dist=0.01; Canberra metric in first UMAP, Euclidean in second
    Selected for qualitatively good-looking clusters and experimentation; the paper states results are not sensitive to these choices but provides no quantitative sensitivity analysis.
  • KDE bandwidth = 0.3 (exponential kernel)
    Chosen because it seemed to recover the distribution of good candidates in the validation set; directly affects the probability densities used for candidate selection.
  • GMM components per band = 200 (L), 200 (S), 50 (C)
    Increased until outlier scores stop depending on the number of Gaussians; this is a hand-tuned model selection.
  • Star-band background cutoff = Candidates from the 5% of star-band combinations with the most candidates are discarded (65% of candidates)
    An ad hoc decision to concentrate effort where RFI contamination is lowest; the paper acknowledges some chance of rejecting a loud real source.
  • Simulation amplitude range = 0 to 4 times the maximum noise in the frequency window
    Set to cover weak to strong signals, but the representativeness of this range for actual ET signals is unverified.
assumptions (5)
  • domain assumption A real technosignature is narrow-band (about Hz) and appears in on-target A observations but not in off-target BCD observations (Sections 1 and 3.1).
    The entire filtering logic and the human vetting definition of a top candidate rely on this cadence-based signature.
  • ad hoc to paper Setigen simulations with the chosen amplitude range and the HIP 17147 S-band background are representative of possible ET signals (Section 3.1).
    No argument establishes that this simulation family spans the unknown signal space; the UMAP/KDE separation is calibrated on it.
  • standard math UMAP embedding preserves enough local structure that clusters in the 2D plane correspond to meaningful signal categories (Sections 3.1 and 3.3).
    UMAP is a heuristic dimensionality-reduction method that distorts global distances; the paper depends on local structure being meaningful.
  • domain assumption Human visual inspection is a reliable gold standard for rejecting RFI and identifying top candidates (Section 4).
    The validation rates are measured against first-stage human labels made primarily by the lead author, and no inter-rater consistency check is reported.
  • domain assumption RFI is the main contaminant and quieter frequency regions are less likely to contain it (Section 3.2).
    The frequency score ranks candidates by inverse density of other candidates, which assumes that local candidate density tracks RFI contamination.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Using anomaly detection to search for technosignatures in Breakthrough Listen observations." pith.science (2026). https://pith.science/paper/X23V34O2

@misc{pith2026250503927,
  author       = {Pith},
  title        = {Pith review of: Using anomaly detection to search for technosignatures in Breakthrough Listen observations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X23V34O2}},
  note         = {Machine review of arXiv:2505.03927}
}
read the original abstract

We implement a machine learning algorithm to search for extra-terrestrial technosignatures in radio observations of several hundred nearby stars, obtained with the Parkes and Green Bank Telescopes by the Breakthrough Listen collaboration. Advances in detection technology have led to an exponential growth in data, necessitating innovative and efficient analysis methods. This problem is exacerbated by the large variety of possible forms an extraterrestrial signal might take, and the size of the multidimensional parameter space that must be searched. It is then made markedly worse by the fact that our best guess at the properties of such a signal is that it might resemble the signals emitted by human technology and communications, the main (yet diverse) contaminant in radio observations. We address this challenge by using a combination of simulations and machine learning methods for anomaly detection. We rank candidates by how unusual they are in frequency, and how persistent they are in time, by measuring the similarity between consecutive spectrograms of the same star. We validate that our filters significantly improve the quality of the candidates that are selected for human vetting when compared to a random selection. Of the ~ 10^11 spectrograms that we analyzed, we visually inspected thousands of the most promising spectrograms, and thousands more for validation, about 20,000 in total, and report that no candidate survived basic scrutiny.

Figures

Figures reproduced from arXiv: 2505.03927 by the authors.

Figure 1
Figure 1. Three examples of observed cadences of the star HIP 66193. The spectrograms on the left show loud RFI throughout the cadence, the middle example shows essentially background noise, and on the right we show what could be a viable candidate based on this view, having a detectable signal only on target. for each pair. This results in 15 unique features (the upper non-diagonal elements of a 6 by 6 matrix) per cadence, r… view at source ↗
Figure 2
Figure 2. Schematic flow chart describing our first filter. We use dimensionality reduction and simulations to identify sources that have a detectable signal on target – in the ‘A’ parts – but not in the alternates, which allows us to reduce the number of candidates by 5 orders of magnitude [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Mapping of simulated signals onto an abstract UMAP space. Every point is a simulated signal from our validation dataset in any or all of the 6 windows of a cadence. In green we mark the signals that are only on-target, while in red we show those that exclusively off target for comparison. All the other clusters in black are various on-off combinations. In light green (red) we delineate the probability derived from t… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Histogram of the probability that a candidate is in the ‘only on-target’ category, using 20 stars. There is an obvious cut-off near 0.0618 which we use as a lower bound to select candidates for further analysis. We examine by eye cadences that ended up in the large clu…
Figure 5
Figure 5. Figure 5: Histogram of all the candidates in each band, after the cross-correlation filter. They seem to prefer some frequencies to others, in a manner that is somewhat correlated with pink bands that are known to be heavily contaminated by RFI, from Price et al. (2019). The gre…
Figure 6
Figure 6. Figure 6: Histogram of the frequency score of all candidates in each band, after the cross-correlation filter. We use these scores to rank candidates for human vetting. We examined and ruled out all the candidates in the top one percent of each band. similarity score for a candi…
Figure 7
Figure 7. Figure 7: Mapping of individual observations of the entire candidates list onto an abstract UMAP space. In green we show a promising candidate, where all the ‘A’ observations seem similar to each other, and different from the off-target ‘B’ observations, which is reflected in a …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 4 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter doi edition editor eprint howpublished institution journal key month number organization pages publisher school series title misctitle type volume year version url label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts ...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION format.url url empty "" new.block "" url * "" * if FUNCTION format.eprint eprint empty "" archivePrefix empty "" archivePrefix "arXiv" = new.block " " eprint * " " * new.block " " eprint * " " * if if if FUNCTION format.doi doi empty "" " " doi * " " * if FUNCTION format.pid doi empty eprint empty ur...

  3. [3]

    ,1B 7Jܰ| Au Myy9g?c1zo v& * 'Y 1ǣ޿ _ @ @ r Vh 1 A L( P . P ח T o ]= a|zh @ q\3㇄ 8

    thebibliography [1] 20pt to REFERENCES 6pt =0pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key on reference command E...

    work page arXiv 2018

  4. [4]

    2017, Monthly Notices, 465, 4530, 10.1093/mnras/stw3021

    Baron, D., & Poznanski, D. 2017, Monthly Notices, 465, 4530, 10.1093/mnras/stw3021

    doi:10.1093/mnras/stw3021 2017

  5. [5]

    P., Croft, S., et al

    Brzycki, B., Siemion, A. P., Croft, S., et al. 2020, Publications of the Astronomical Society of the Pacific, 132, 114501, 10.1088/1538-3873/ABAAF7

    doi:10.1088/1538-3873/abaaf7 2020

  6. [6]

    Brzycki, B., Siemion, A. P. V., Pater, I. D., et al. 2022, The Astronomical Journal, 164, 33, 10.3847/1538-3881/ac5e3d

    doi:10.3847/1538-3881/ac5e3d 2022

  7. [7]

    2024, , 167, 10, 10.3847/1538-3881/acf576

    Choza , C., Bautista , D., Croft , S., et al. 2024, , 167, 10, 10.3847/1538-3881/acf576

    doi:10.3847/1538-3881/acf576 2024

  8. [8]

    A., Egly, S., Harp, G

    Cox, G. A., Egly, S., Harp, G. R., et al. 2018, Proceedings of the IEEE Aerospace Conference, 1, 10.48550/arxiv.1803.08624

    arXiv doi:10.48550/arxiv.1803.08624 2018

Show all 40 references
  1. [9]

    2021, , 133, 064502, 10.1088/1538-3873/abf329

    Czech , D., Isaacson , H., Pearce , L., et al. 2021, , 133, 064502, 10.1088/1538-3873/abf329

    2021 doi

  2. [10]

    2001, A Practical Guide to Splines - Revised Edition, Vol

    de Boor, C. 2001, A Practical Guide to Splines - Revised Edition, Vol. 115 (Springer), 366. https://link.springer.com/book/9780387953663

    2001

  3. [11]

    Drake, F. D. 1961, Citation: Physics Today, 14, 48, 10.1063/1.3057500

    1961 doi

  4. [12]

    2019, turboSETI: Python-based SETI search algorithm , Astrophysics Source Code Library, record ascl:1906.006

    Enriquez , E., & Price , D. 2019, turboSETI: Python-based SETI search algorithm , Astrophysics Source Code Library, record ascl:1906.006

    2019

  5. [13]

    E., Siemion, A., Foster, G., et al

    Enriquez, J. E., Siemion, A., Foster, G., et al. 2017, The Astrophysical Journal, 849, 104, 10.3847/1538-4357/aa8d1b

    2017 doi

  6. [14]

    Franz, N., Croft, S., Siemion, A. P. V., et al. 2022, The Astronomical Journal, 163, 104, 10.3847/1538-3881/ac46c9

    2022 doi

  7. [15]

    I., Siemion, A

    Gajjar, V., Perez, K. I., Siemion, A. P. V., et al. 2021, The Astronomical Journal, 162, 33, 10.3847/1538-3881/abfd36

    2021 doi

  8. [16]

    R., Richards, J., Shostak, S., et al

    Harp, G. R., Richards, J., Shostak, S., et al. 2019, arXiv, 10.48550/arxiv.1902.02426

    2019 doi arXiv

  9. [17]

    A., & Yawney, J

    Hilal, W., Gadsden, S. A., & Yawney, J. 2022, Expert Systems with Applications, 193, 116429, https://doi.org/10.1016/j.eswa.2021.116429

    2022

  10. [18]

    2024, arXiv e-prints, arXiv:2411.16556, 10.48550/arXiv.2411.16556

    Jacobson-Bell , B., Croft , S., Choza , C., et al. 2024, arXiv e-prints, arXiv:2411.16556, 10.48550/arXiv.2411.16556

    2024 doi arXiv

  11. [19]

    Lebofsky, M., Croft, S., Siemion, A. P. V., et al. 2019, Publications of the Astronomical Society of the Pacific, 131, 10.1088/1538-3873/ab3e82

    2019 doi

  12. [20]

    X., Croft, S., Lintott, C., & Siemion, A

    Ma, P. X., Croft, S., Lintott, C., & Siemion, A. P. V. 2024, RAS Techniques and Instruments, 3, 33, 10.1093/rasti/rzad056

    2024 doi

  13. [21]

    X., Ng, C., Rizk, L., et al

    Ma, P. X., Ng, C., Rizk, L., et al. 2023, Nature Astronomy, 10.1038/s41550-022-01872-z

    2023 doi

  14. [22]

    H., Price, D

    Macmahon, D. H., Price, D. C., Lebofsky, M., et al. 2018, Publications of the Astronomical Society of the Pacific, 130, 044502, 10.1088/1538-3873/AA80D2

    2018 doi

  15. [23]

    2020, The Astronomical Journal, 161, 55, 10.3847/1538-3881/abcc77

    Margot, J.-L., Pinchuk, P., Geil, R., et al. 2020, The Astronomical Journal, 161, 55, 10.3847/1538-3881/abcc77

    2020 doi

  16. [24]

    G., Pinchuk , P., et al

    Margot , J.-L., Li , M. G., Pinchuk , P., et al. 2023, , 166, 206, 10.3847/1538-3881/acfda4

    2023 doi

  17. [25]

    2018, Journal of Open Source Software, 3, 861, 10.48550/arxiv.1802.03426

    McInnes, L., Healy, J., & Melville, J. 2018, Journal of Open Source Software, 3, 861, 10.48550/arxiv.1802.03426

    2018 doi arXiv

  18. [26]

    A., Pota, M., & Esposito, M

    Panza, M. A., Pota, M., & Esposito, M. 2023, Electronics, 12, 10.3390/electronics12183971

    2023 doi

  19. [27]

    2011, Journal of Machine Learning Research, 12, 2825, 10.5555/1953048

    Pedregosa, F., Michel, V., Grisel, O., et al. 2011, Journal of Machine Learning Research, 12, 2825, 10.5555/1953048

    2011 doi

  20. [28]

    2022, The Astronomical Journal, 163, 76, 10.3847/1538-3881/AC426F

    Pinchuk, P., & Margot, J.-L. 2022, The Astronomical Journal, 163, 76, 10.3847/1538-3881/AC426F

    2022 doi

  21. [29]

    H., et al

    Pinchuk, P., Margot, J.-L., Greenberg, A. H., et al. 2019, The Astronomical Journal, 157, 122, 10.3847/1538-3881/AB0105

    2019 doi

  22. [30]

    C., Enriquez, J

    Price, D. C., Enriquez, J. E., Brzycki, B., et al. 2019, The Astronomical Journal, 159, 86, 10.3847/1538-3881/ab65f1

    2019 doi

  23. [31]

    2018, , 476, 2117, 10.1093/mnras/sty348

    Reis , I., Poznanski , D., Baron , D., Zasowski , G., & Shahaf , S. 2018, , 476, 2117, 10.1093/mnras/sty348

    2018 doi

  24. [32]

    X., & Wolf , L

    Reis , I., Rotman , M., Poznanski , D., Prochaska , J. X., & Wolf , L. 2021, Astronomy and Computing, 34, 100437, 10.1016/j.ascom.2020.100437

    2021

  25. [33]

    R., Vandermeulen, R

    Ruff, L., Kauffmann, J. R., Vandermeulen, R. A., et al. 2021, Proceedings of the IEEE, 109, 756–795, 10.1109/jproc.2021.3052449

    2021

  26. [34]

    J., & Vanderburg , A

    Shallue , C. J., & Vanderburg , A. 2018, , 155, 94, 10.3847/1538-3881/aa9e09

    2018 doi

  27. [35]

    P., Demorest, P., Korpela, E., et al

    Siemion, A. P., Demorest, P., Korpela, E., et al. 2013, Astrophysical Journal, 767, 94, 10.1088/0004-637X/767/1/94

    2013 doi

  28. [36]

    Siemion, A. P. V., Bower, G. C., Foster, G., et al. 2011, Astrophysical Journal, 744, 10.1088/0004-637X/744/2/109

    2011 doi

  29. [37]

    E., & Gadermayr, M

    Tschuchnig, M. E., & Gadermayr, M. 2022, Anomaly Detection in Medical Imaging - A Mini Review (Springer Fachmedien Wiesbaden), 33–38, 10.1007/978-3-658-36295-9_5

    2022 doi

  30. [38]

    2008, Journal of Machine Learning Research, 9, 2579

    van der Maaten, L., & Hinton, G. 2008, Journal of Machine Learning Research, 9, 2579. https://www.jmlr.org/papers/v9/vandermaaten08a.html

    2008

  31. [39]

    A., Cranmer, M., Berger, E., et al

    Villar, V. A., Cranmer, M., Berger, E., et al. 2021, The Astrophysical Journal Supplement Series, 255, 24, 10.3847/1538-4365/ac0893

    2021 doi

  32. [40]

    G., Won, K

    Zhang, Y. G., Won, K. H., Son, S. W., Siemion, A., & Croft, S. 2019, 2018 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2018 - Proceedings, 1114, 10.1109/GLOBALSIP.2018.8646437

    2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.