Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Representation learning for fast radio burst dynamic spectra

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read An IOB-augmented convolutional autoencoder reconstructs complex FRB dynamic spectra better than PCA using fewer latent variables.

desk verdict A solid, honest methods comparison with a shareable simulation tool; the IOB-CAE advantage is plausible but needs error bars and a signal-region metric to be fully convincing. read the letter →

arxiv 2412.12394 v2 pith:KGMXEMFJ submitted 2024-12-16 astro-ph.HE

Dirk Kuiper , Gabriella Contardo , Daniela Huppenkothen , Jason W. T. Hessels This is my paper · ORCID
classification astro-ph.HE
keywords fastradioburstsdynamicspectrarepresentationlearningconvolutionalautoencoderinformation-orderedbottleneckprincipalcomponentanalysisunsupervisedclassificationFRBmorphology
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 tries to establish that unsupervised representation learning can capture the time–frequency morphology of fast radio bursts directly from their dynamic spectra, without reducing each burst to hand-picked summary parameters. It claims that a convolutional autoencoder with an Information-Ordered Bottleneck (IOB) layer reconstructs simulated FRB dynamic spectra—including scattered, complex, and downward-drifting bursts—with lower mean-squared error and far fewer latent variables than Principal Component Analysis (PCA). The same model generalizes to real CHIME bursts and places them in a latent space where simple classes partially separate while complex morphologies form a continuum. If correct, this would give astronomers a compact, data-driven coordinate system for organizing the rapidly growing FRB population and for spotting physically unusual bursts.

What carries the argument

The Information-Ordered Bottleneck (IOB) is the mechanism that carries the argument: during training it varies the number of active latent variables by masking, forcing the network to pass information through the first k units and to order units by their contribution to reconstruction. Placed between the encoder and decoder of a convolutional autoencoder, it produces a compact, ordered latent space that preserves essential morphology while discarding noise. The companion FRBakery simulation tool supplies a controlled, five-class dataset (simple broad, simple narrow, scattered, complex, drifting) used to measure reconstruction performance and to interpret the latent space.

What would settle it

Compare the IOB-CAE and PCA on a held-out sample of real high-time-resolution CHIME bursts with morphology labels: if the IOB-CAE does not reconstruct scattered and drifting bursts with lower MSE than PCA at fourteen components, the central claim fails. Alternatively, if an expanded, unbiased real dataset yields well-separated morphology clusters in the IOB latent space instead of a continuum, the continuum claim fails.

Watch

Extended reading notes

Core claim

The central claim is that the IOB-augmented convolutional autoencoder learns a low-dimensional representation in which reconstruction error drops steeply and plateaus at roughly six to eight latent variables, capturing the intricate structure of scattered, complex, and drifting bursts that PCA at fourteen components cannot reproduce. Applied to a combined training set of FRBakery simulations and real CHIME complex-voltage Stokes I spectra, the encoder places simple narrow and broad bursts in partially separated regions, while scattered, complex, and drifting bursts overlap along a morphological continuum. PCA, run on the real CHIME sample alone, flags a set of outliers whose dynamic spectra show multiple components, scattering tails, or instrumental channelization artifacts. The paper frames these results as evidence that FRBs may not form discrete clusters, but that learned representations can still uncover meaningful structure in their diversity.

Load-bearing premise

The load-bearing premise is that the FRBakery simulations faithfully represent real FRB morphologies in their parameter ranges, scattering shapes, and class balance; if the real FRB population differs in any of these, the measured reconstruction advantage and the latent-space interpretation may not transfer.

Editorial extensions

If this is right

  • If the claim holds, roughly six to eight ordered latent variables are enough to encode both global and fine-scale structure of FRB dynamic spectra, making large-catalog morphological studies computationally feasible.
  • Reconstruction comparisons imply that PCA systematically underestimates the structure of scattered, complex, and drifting bursts, so linear summaries of such bursts should be treated with caution.
  • The observed latent-space continuum suggests that FRB morphology is better described by continuous variation than by discrete classes, which would reframe classification efforts around distances or trajectories in the latent space.
  • PCA outlier detection on real data can surface both astrophysically interesting bursts and instrumental artifacts, providing a screening step for future surveys.
  • Because the model trains on about five thousand simulated bursts and still reconstructs real CHIME bursts, the approach is scalable to the much larger samples expected from next-generation telescopes.

Reading between the lines

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

  • If the latent continuum is physically real, the repeater/non-repeater dichotomy may be endpoints of a continuous morphology axis rather than a clean split; one could test this by regressing latent coordinates against scattering time and drift rate.
  • The IOB's ordering property is itself a testable feature selector: the first few latent variables could be compared directly with physical observables to see whether the network rediscovers quantities like burst duration or frequency bandwidth.
  • The MSE comparison is dominated by noise-dominated pixels, so the apparent advantage of the IOB-CAE might shrink or grow under a signal-weighted loss; an SSIM-based loss is a natural follow-up the paper itself mentions.
  • A stronger test of generalization would be to train on an unbalanced, realistic class distribution and check whether the latent-space continuum persists or is an artifact of the balanced simulation prior.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper presents a comparison of two unsupervised dimensionality-reduction methods for fast radio burst (FRB) dynamic spectra: principal component analysis (PCA) and a convolutional autoencoder augmented with an information-ordered bottleneck (IOB-CAE). The authors introduce FRBakery, a simulation tool that generates synthetic FRB dynamic spectra in five morphology classes (simple narrow, simple broad, scattered, complex, and drifting), train both methods on these simulations, and validate on CHIME complex-voltage bursts. The central claims are that the IOB-CAE reconstructs complex FRB structures with lower mean squared error and fewer latent variables than PCA, that it denoises effectively at modest signal-to-noise ratios, and that its latent space organizes real CHIME bursts into a continuum of morphologies with only partial class separation.

Significance. If these conclusions hold, the paper offers a useful, scalable framework for morphology-oriented FRB population studies and demonstrates a sensible application of the IOB architecture to astronomical imaging data. The authors make their simulation and analysis code openly available, provide machine-readable data availability statements, and ground the work in a non-trivial combination of simulated and real observations. The IOB's adaptive bottleneck is a principled approach to learning ordered latent representations, and the paper's emphasis on representation quality rather than supervised classification is timely. However, the quantitative support for the strongest claims is currently fragile: the MSE comparisons lack statistical grounding, the real-data evaluation is mostly qualitative, and the latent-space continuum conclusion rests heavily on hand-selected simulated parameter ranges. The work is publishable in principle, but the load-bearing numerical evidence needs to be strengthened.

major comments (4)
  1. [Section 4, Figure 5] The MSE comparisons in Figure 5 are presented without error bars, repeated training runs, or significance tests. The paper's central quantitative claim that the IOB-CAE 'significantly outperforms' PCA in reconstruction quality therefore rests on point estimates from a single train/test split and a single model initialization. Please report the mean and standard deviation of the MSE across multiple random seeds and, ideally, paired per-burst differences between the two methods to establish that the observed advantage is not run-to-run scatter.
  2. [Section 4, reconstruction loss discussion] The paper itself notes that noise-only bins outnumber signal bins by over an order of magnitude, and the MSE is computed on standardized dynamic spectra. In this regime the global MSE can be dominated by how each method treats the background rather than by how well it preserves signal structure: a nonlinear autoencoder can learn to output a near-mean background in noise-dominated regions, while PCA with few components will necessarily reconstruct some of the noise variance. Because the authors do not separate signal-region from noise-region error, or report a structure-sensitive metric such as SSIM, the conclusion that the IOB-CAE better captures 'nuanced, non-linear structures' is not yet established. Please quantify the reconstruction error separately in signal-dominated and noise-dominated pixels, and/or report a perceptual or structural metric.
  3. [Section 5.1 and 5.3] The real-data evaluation is qualitative: Figures 7 and 11 show selected latent-space projections and reconstructions, but there are no quantitative metrics for denoising, generalization, or class separation on the CHIME bursts. The abstract's claim of 'effective denoising at modest signal-to-noise ratios' is therefore unsupported on real data. Please provide quantitative evaluation on held-out real bursts, for example reconstruction error in signal regions, an injection-recovery test at known S/N levels, or a clustering/separation metric on the latent-space coordinates. In addition, please state explicitly whether the CHIME points shown in Figures 7, 8, and 11 are from the training or test split; if they are from the training set, the generalization claim in Section 5.3 is weakened.
  4. [Section 2.1 and Table 1] The simulated training population is evenly balanced across five hand-selected morphology classes, with parameter ranges informed by observed trends but not by a quantitative population model. The claim in Section 5.1 that FRBs form a morphological continuum may partly reflect this training distribution rather than the intrinsic structure of real bursts. This is not circularity in the strict sense, but its robustness needs testing: please train on class-imbalanced or reweighted simulated sets, vary the parameter ranges, and check whether the latent-space continuum persists. The small subset of real CHIME bursts currently used cannot independently establish the continuum claim.
minor comments (4)
  1. [Section 2.1, Table 1] There are typos in Table 1 and the surrounding text: 'S/N scaling exponentt' should be 'exponent', and 'WILL' should be 'wll' or properly cited as a software package.
  2. [Section 4] There are several typographical errors, including 'reconstructin', 'effectively reconstruct..', and an extra period at the end of a sentence in the discussion of the MSE plateau. A careful proofreading pass is needed.
  3. [Figures 5 and 6] The axis labels are incomplete: the MSE curves should specify whether the error is computed on standardized flux units, and the x-axis of Figure 5b should be labeled as the number of latent variables/bottleneck width. Adding these details would improve reproducibility.
  4. [Section 6.2] The outlook lists SSIM as a possible future loss function, but given the noise-dominated MSE concern raised in Section 4, evaluating such a structure-sensitive metric in the present study would directly address the manuscript's central quantitative claim rather than being deferred to future work.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: reconstruction performance is measured on held-out simulated data, and the IOB method is an externally cited architecture; acknowledged limitations concern metric validity, not self-referential derivation.

full rationale

The paper's central claims are not equivalent to their inputs by construction. The reconstruction comparison (Section 4, Figures 5-6) is computed on a held-out 20% test set from the FRBakery simulations: PCA components are derived from the training set and the IOB-CAE is trained on the training set, so the MSE curves measure out-of-sample generalization rather than a fitted value being renamed as a prediction. The IOB layer is adopted from Ho et al. (2024), an external reference with no author overlap, and no uniqueness theorem or load-bearing self-citation is invoked. The latent-space analysis of real CHIME bursts (Section 5.1) is an empirical projection of data through a model trained on a combined set; although the simulation parameter ranges are hand-selected and evenly balanced, the paper explicitly cautions that the synthetic set is not representative of the true FRB population and that the observed continuum may reflect pixel-space geometry as well as intrinsic diversity. The manuscript's own caveat that noise-dominated bins can dominate MSE is a limitation of the chosen metric, not a circular reduction: the metric is the same for both methods, and the qualitative reconstructions provide independent (if selected) evidence. No step in the derivation chain reduces to its inputs by definition, so the analysis is self-contained against external benchmarks and the circularity score is 0.

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

The central validation depends on hand-selected simulation parameters and several domain assumptions about the faithfulness of the synthetic data and preprocessing; there are no invented physical entities, and the IOB layer itself is taken from prior work.

free parameters (6)
  • S/N power-law exponent = -1.5
    Used in FRBakery to sample burst signal-to-noise ratios for all five classes (Table 1). Chosen by hand to approximate observed brightness distributions.
  • Time width distributions = 0.0008 +/- 0.0004 s (SN/SB), 0.0004 +/- 0.0002 s (SC/CP), 0.0002 +/- 0.0001 s (DD)
    Hand-selected ranges in Table 1 based on observed FRB durations; they shape how separable the classes are in the latent space.
  • Frequency width distributions = 12.5-100 MHz (narrow classes), 150-200 MHz (simple broad)
    Hand-selected ranges in Table 1; the broad-versus-narrow boundary is arbitrary and contributes to the observed continuum.
  • Scattering time = 0.001 +/- 0.001 s
    Hand-selected for the Scattered class (Table 1) based on observed scattering timescales.
  • Drift rate = 100-200 MHz/ms
    Hand-selected for the Drifting class (Table 1) to mimic 'sad trombone' bursts.
  • CAE architecture hyperparameters = kernel 3x3, stride 2, padding 1, two conv layers, latent size not fixed; see Figure 4
    Selected through iterative experimentation (Section 3.2.5) rather than a systematic search; performance claims depend on this choice.
assumptions (6)
  • standard math PCA eigendecomposition of the sample covariance gives the optimal linear reconstruction.
    Invoked in Section 3.1 to define principal components and explained variance.
  • standard math Adam with early stopping converges to a well-generalizing autoencoder.
    Training procedure in Section 3.2.6; no proof that the found minimum is the best, but standard practice.
  • domain assumption The five morphology classes (simple broad, simple narrow, scattered, complex, drifting) span the relevant diversity of FRB dynamic spectra.
    Section 2.1; if real FRBs contain morphologies outside these classes, the validation is incomplete.
  • domain assumption CHIME complex voltage preprocessing, including padding/truncation to 976x1024 and filling zapped channels with Gaussian noise, preserves morphology-relevant information.
    Section 2.2; these transformations could distort or inject features.
  • domain assumption Distance in the latent space approximates morphological similarity.
    Section 5.1; used to interpret Figure 7 and the continuum claim.
  • ad hoc to paper Complex bursts appear in the latent bulk because they are reconstructed less accurately.
    Section 5.1; a post hoc interpretation offered without direct evidence linking reconstruction error to latent position.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Representation learning for fast radio burst dynamic spectra." pith.science (2026). https://pith.science/paper/KGMXEMFJ

@misc{pith2026241212394,
  author       = {Pith},
  title        = {Pith review of: Representation learning for fast radio burst dynamic spectra},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KGMXEMFJ}},
  note         = {Machine review of arXiv:2412.12394}
}
read the original abstract

Fast radio bursts (FRBs) are millisecond-duration radio transients of extragalactic origin, with diverse time-frequency patterns and emission properties that require explanation. With one possible exception, FRBs are detected only in the radio, so analyzing their dynamic spectra is therefore crucial to disentangling the physical processes governing their generation and propagation. Furthermore, comparing FRB morphologies provides insights into possible differences among their progenitors and environments. This study applies unsupervised learning and deep learning techniques to investigate FRB dynamic spectra, focusing on two approaches: Principal Component Analysis (PCA) and a Convolutional Autoencoder (CAE) enhanced by an Information-Ordered Bottleneck (IOB) layer. PCA served as a computationally efficient baseline, capturing broad trends, identifying outliers, and providing valuable insights into large datasets. However, its linear nature limited its ability to reconstruct complex FRB structures. In contrast, the IOB-augmented CAE excelled at capturing intricate features, with high reconstruction accuracy and effective denoising at modest signal-to-noise ratios. The IOB layer's ability to prioritize relevant features enabled efficient data compression, preserving key morphological characteristics with minimal latent variables. When applied to real FRBs from CHIME, the IOB-CAE generalized effectively, revealing a latent space that highlighted the continuum of FRB morphologies and the potential for distinguishing intrinsic differences between burst types. This framework demonstrates that while FRBs may not naturally cluster into discrete groups, advanced representation learning techniques can uncover meaningful structures, offering new insights into the diversity and origins of these bursts.

Figures

Figures reproduced from arXiv: 2412.12394 by the authors.

Figure 1
Figure 1. Dynamic spectra with corresponding lightcurves (averaged across all frequencies), for selected bursts that represent different morphological classes. Each row represents a burst category (Simple Broad, Simple Narrow, Scattered, Complex, and Drifting, respectively). Each column shows individual burst examples within that category. Above each dynamic spectrum, the black line represents the lightcurve of the burst. The… view at source ↗
Figure 2
Figure 2. Dynamic spectrum of processed complex voltage data from the CHIME telescope plotting the total intensity (Stokes I). The bottom panel shows the dynamic spectrum of the FRB signal, where each row represents a frequency channel over time. Zapped channels, filled with characteristic noise, are marked with red ticks on the y-axis. The top panel displays the integrated intensity time series, summing the signal across all… view at source ↗
Figure 3
Figure 3. Conceptual design of the IOB. At each training step, the bottleneck width 𝑘 is varied by masking inactive latent variables. The active variables are prioritized by their contribution to the reconstruction, creating a structured and ordered latent space. Figure taken from Ho et al. (2024). is achieved by dynamically varying the size of the bottleneck during training and ordering the latent variables by their contribu… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Structure of the Convolutional Autoencoder (CAE) with the Information Ordered Bottleneck (IOB) layer. The encoder compresses the input data, the IOB layer refines the latent representation by adaptively ordering latent variables, and the decoder reconstructs the input …
Figure 5
Figure 5. Figure 5: Comparison of reconstruction performance for PCA and IOB-CAE. The left panel shows the MSE as a function of the number of PCA components, with decreasing MSE indicating improved reconstruction accuracy. The right panel illustrates MSE as a function of bottleneck width …
Figure 6
Figure 6. Figure 6: Comparison of PCA and IOB-CAE reconstructions for different FRB burst classes (Simple Broad, Simple Narrow, Scattered, Complex, and Drifting). Each row corresponds to a burst category, with three columns displaying the original burst, the reconstructed burst, and the d…
Figure 7
Figure 7. Figure 7: First five components of the latent space for the convolutional autoencoder with IOB using both simulated and real CHIME bursts. Different colors represent different classes. 5 EXTENDING TO REAL DATA 5.1 Latent Space Representation To evaluate how well the IOB-CAE and …
Figure 8
Figure 8. Figure 8: First five components of PCA using both simulated and real CHIME bursts. Different colors represent different classes. astrophysical burst classification, enabling systematic identification of unusual bursts and potential instrumental effects. 5.3 IOB-CAE Reconstructio…
Figure 9
Figure 9. Figure 9: PCA visualization of the CHIME bursts, highlighting multiple outliers. The first two principal components are shown, with the most significant outliers labeled and color-coded for further analysis. Many of these outliers exhibit complex burst structures, such as multip…
Figure 10
Figure 10. Figure 10: Dynamic spectra of the nine identified outliers from the CHIME dataset, showing unusual spectral characteristics. Some of these bursts exhibit complex morphology, including multiple components or pronounced scattering tails, while others, such as FRB 20190417C and FRB…
Figure 11
Figure 11. Figure 11: Reconstruction quality of IOB-CAE using real CHIME bursts. Each row represents a different burst type (Simple Broad, Simple Narrow, Scattered, Complex, Drifting and CHIME). The first column shows the original CHIME bursts, while each subsequent column shows reconstruc…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Semi-supervised morphological classification of fast radio bursts from the second CHIME/FRB catalogue

    astro-ph.HE 2026-07 conditional novelty 5.0 of 10

    A simulation-trained convolutional autoencoder identifies three distinct FRB morphological classes and classifies repeatability with 86% recall.

Reference graph

Works this paper leans on

77 extracted references · 15 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTION or pop #1...

  2. [2]

    Agarwal D., et al., 2020, @doi [ ] 10.1093/mnras/staa1927 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.497..352A 497, 352

    work page doi:10.1093/mnras/staa1927 2020

  3. [3]

    Amiri M., et al., 2021, @doi [The Astrophysical Journal Supplement Series] 10.3847/1538-4365/ac33ab , 257, 59

    doi:10.3847/1538-4365/ac33ab 2021

  4. [4]

    J.] 10.3847/1538-4357/ad464b , 969, 145

    Amiri M., et al., 2024, @doi [Astrophys. J.] 10.3847/1538-4357/ad464b , 969, 145

    doi:10.3847/1538-4357/ad464b 2024

  5. [5]

    Andersen B., et al., 2020, @doi [Nature] 10.1038/s41586-020-2863-y , 587, 54–58

    doi:10.1038/s41586-020-2863-y 2020

  6. [6]

    W., et al., 2019, @doi [Science] 10.1126/science.aaw5903 , https://ui.adsabs.harvard.edu/abs/2019Sci...365..565B 365, 565

    Bannister K. W., et al., 2019, @doi [Science] 10.1126/science.aaw5903 , https://ui.adsabs.harvard.edu/abs/2019Sci...365..565B 365, 565

    doi:10.1126/science.aaw5903 2019

  7. [7]

    Bhandari S., et al., 2018, @doi [ ] 10.1093/mnras/stx3074 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.475.1427B 475, 1427

    doi:10.1093/mnras/stx3074 2018

  8. [8]

    D., Ravi V., Belov K

    Bochenek C. D., Ravi V., Belov K. V., Hallinan G., Kocz J., Kulkarni S. R., McKenna D. L., 2020, @doi [ ] 10.1038/s41586-020-2872-x , https://ui.adsabs.harvard.edu/abs/2020Natur.587...59B 587, 59

    doi:10.1038/s41586-020-2872-x 2020

Show all 77 references
  1. [9]

    CHIME/FRB Collaboration et al., 2019, @doi [ ] 10.1038/s41586-018-0867-7 , https://ui.adsabs.harvard.edu/abs/2019Natur.566..230C 566, 230

    2019 doi

  2. [10]

    CHIME/FRB Collaboration B. C. A., et al., 2022, @doi [Nature] 10.1038/s41586-022-04841-8 , 607, 256

    2022 doi

  3. [11]

    J., et al., 2016, @doi [ ] 10.1093/mnrasl/slw069 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.460L..30C 460, L30

    Champion D. J., et al., 2016, @doi [ ] 10.1093/mnrasl/slw069 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.460L..30C 460, L30

    2016 doi

  4. [12]

    Chen G., Ravi V., Lu W., 2020, @doi [ ] 10.3847/1538-4357/ab982b , https://ui.adsabs.harvard.edu/abs/2020ApJ...897..146C 897, 146

    2020 doi

  5. [13]

    H., Hashimoto T., Goto T., Kim S

    Chen B. H., Hashimoto T., Goto T., Kim S. J., Santos D. J. D., On A. Y. L., Lu T.-Y., Hsiao T. Y.-Y., 2021, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stab2994 , 509, 1227

    2021 doi

  6. [14]

    Y., Annex A., Ali-Dib M., Szabo R., Jackman C

    Chen T. Y., Annex A., Ali-Dib M., Szabo R., Jackman C. M., Amerstorfer U. V., B \' o di A., 2023, Bulletin of the AAS, 55

    2023

  7. [15]

    Connor L., Petroff E., 2018, @doi [ ] 10.3847/2041-8213/aacd02 , https://ui.adsabs.harvard.edu/abs/2018ApJ...861L...1C 861, L1

    2018 doi

  8. [16]

    M., Chatterjee S., 2019, @doi [Annual Review of Astronomy and Astrophysics] 10.1146/annurev-astro-091918-104501 , 57, 417–465

    Cordes J. M., Chatterjee S., 2019, @doi [Annual Review of Astronomy and Astrophysics] 10.1146/annurev-astro-091918-104501 , 57, 417–465

    2019 doi

  9. [17]

    Damiano M., Micela G., Tinetti G., 2019, @doi [ ] 10.3847/1538-4357/ab22b2 , https://ui.adsabs.harvard.edu/abs/2019ApJ...878..153D 878, 153

    2019 doi

  10. [18]

    K., et al., 2020, @doi [ ] 10.1093/mnras/staa2138 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.497.3335D 497, 3335

    Day C. K., et al., 2020, @doi [ ] 10.1093/mnras/staa2138 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.497.3335D 497, 3335

    2020 doi

  11. [19]

    T., et al., 2024, @doi [ ] 10.3847/1538-4357/ad59aa , https://ui.adsabs.harvard.edu/abs/2024ApJ...974..274F 974, 274

    Faber J. T., et al., 2024, @doi [ ] 10.3847/1538-4357/ad59aa , https://ui.adsabs.harvard.edu/abs/2024ApJ...974..274F 974, 274

    2024 doi

  12. [20]

    Farah W., et al., 2018, @doi [ ] 10.1093/mnras/sty1122 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.478.1209F 478, 1209

    2018 doi

  13. [21]

    Hessels J. W. T., et al., 2019, @doi [ ] 10.3847/2041-8213/ab13ae , https://ui.adsabs.harvard.edu/abs/2019ApJ...876L..23H 876, L23

    2019 doi

  14. [22]

    M., et al., 2023, @doi [ ] 10.1093/mnras/stad2847 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.526.2039H 526, 2039

    Hewitt D. M., et al., 2023, @doi [ ] 10.1093/mnras/stad2847 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.526.2039H 526, 2039

    2023 doi

  15. [23]

    H., Peter Schloerb F., 1997, @doi [ ] 10.1086/303514 , https://ui.adsabs.harvard.edu/abs/1997ApJ...475..173H 475, 173

    Heyer M. H., Peter Schloerb F., 1997, @doi [ ] 10.1086/303514 , https://ui.adsabs.harvard.edu/abs/1997ApJ...475..173H 475, 173

    1997 doi

  16. [24]

    E., Salakhutdinov R

    Hinton G. E., Salakhutdinov R. R., 2006, @doi [Science] 10.1126/science.1127647 , https://ui.adsabs.harvard.edu/abs/2006Sci...313..504H 313, 504

    2006 doi

  17. [25]

    D., 2024, Information-Ordered Bottlenecks for Adaptive Dimensionality Reduction, https://openreview.net/forum?id=XbydvPq92M

    Ho M., Zhao X., Wandelt B. D., 2024, Information-Ordered Bottlenecks for Adaptive Dimensionality Reduction, https://openreview.net/forum?id=XbydvPq92M

    2024

  18. [26]

    Springer Series in Statistics, Springer, https://books.google.nl/books?id=_olByCrhjwIC

    Jolliffe I., 2002, Principal Component Analysis. Springer Series in Statistics, Springer, https://books.google.nl/books?id=_olByCrhjwIC

    2002

  19. [27]

    W., 2023, Will, https://github.com/josephwkania/will

    Kania J. W., 2023, Will, https://github.com/josephwkania/will

    2023

  20. [28]

    P., Ba J., 2017, Adam: A Method for Stochastic Optimization ( @eprint arXiv 1412.6980 )

    Kingma D. P., Ba J., 2017, Adam: A Method for Stochastic Optimization ( @eprint arXiv 1412.6980 )

    2017 arXiv

  21. [29]

    P., Welling M., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1906.02691 , https://ui.adsabs.harvard.edu/abs/2019arXiv190602691K p

    Kingma D. P., Welling M., 2019, @doi [arXiv e-prints] 10.48550/arXiv.1906.02691 , https://ui.adsabs.harvard.edu/abs/2019arXiv190602691K p. arXiv:1906.02691

    2019 doi arXiv

  22. [30]

    Kirsten F., et al., 2022, @doi [ ] 10.1038/s41586-021-04354-w , https://ui.adsabs.harvard.edu/abs/2022Natur.602..585K 602, 585

    2022 doi

  23. [31]

    Kirsten F., et al., 2024, @doi [Nature Astronomy] 10.1038/s41550-023-02153-z , https://ui.adsabs.harvard.edu/abs/2024NatAs...8..337K 8, 337

    2024 doi

  24. [32]

    W., 2024, @doi [Nature Astronomy] 10.1038/s41550-023-02125-3 , https://ui.adsabs.harvard.edu/abs/2024NatAs...8..230K 8, 230

    Kramer M., Liu K., Desvignes G., Karuppusamy R., Stappers B. W., 2024, @doi [Nature Astronomy] 10.1038/s41550-023-02125-3 , https://ui.adsabs.harvard.edu/abs/2024NatAs...8..230K 8, 230

    2024 doi

  25. [33]

    E., 2012, -, 25

    Krizhevsky A., Sutskever I., Hinton G. E., 2012, -, 25

    2012

  26. [34]

    LeCun Y., Bengio Y., Hinton G., 2015, @doi [ ] 10.1038/nature14539 , https://ui.adsabs.harvard.edu/abs/2015Natur.521..436L 521, 436

    2015 doi

  27. [35]

    Lin H.-H., et al., 2022, @doi [Publications of the Astronomical Society of the Pacific] 10.1088/1538-3873/ac8f71 , 134, 094106

    2022 doi

  28. [36]

    R., Bailes M., McLaughlin M

    Lorimer D. R., Bailes M., McLaughlin M. A., Narkevic D. J., Crawford F., 2007, @doi [Science] 10.1126/science.1147532 , https://ui.adsabs.harvard.edu/abs/2007Sci...318..777L 318, 777

    2007 doi

  29. [37]

    Lucie-Smith L., Despali G., Springel V., 2024, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stae1487 , 532, 164

    2024 doi

  30. [38]

    Luo J.-W., Zhu-Ge J.-M., Zhang B., 2023, @doi [ ] 10.1093/mnras/stac3206 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.518.1629L 518, 1629

    2023 doi

  31. [39]

    P., et al., 2015, in Advancing Astrophysics with the Square Kilometre Array (AASKA14)

    Macquart J. P., et al., 2015, in Advancing Astrophysics with the Square Kilometre Array (AASKA14). p. 55 ( @eprint arXiv 1501.07535 ), @doi 10.22323/1.215.0055

    2015 arXiv

  32. [40]

    P., et al., 2020, @doi [ ] 10.1038/s41586-020-2300-2 , https://ui.adsabs.harvard.edu/abs/2020Natur.581..391M 581, 391

    Macquart J. P., et al., 2020, @doi [ ] 10.1038/s41586-020-2300-2 , https://ui.adsabs.harvard.edu/abs/2020Natur.581..391M 581, 391

    2020 doi

  33. [41]

    A., et al., 2021, @doi [ ] 10.3847/2041-8213/ac1921 , https://ui.adsabs.harvard.edu/abs/2021ApJ...919L...6M 919, L6

    Majid W. A., et al., 2021, @doi [ ] 10.3847/2041-8213/ac1921 , https://ui.adsabs.harvard.edu/abs/2021ApJ...919L...6M 919, L6

    2021 doi

  34. [42]

    Springer Berlin Heidelberg, Berlin, Heidelberg, pp 52--59

    Masci J., Meier U., Cire s an D., Schmidhuber J., 2011, in Honkela T., Duch W., Girolami M., Kaski S., eds, Artificial Neural Networks and Machine Learning -- ICANN 2011. Springer Berlin Heidelberg, Berlin, Heidelberg, pp 52--59

    2011

  35. [43]

    McInnes L., Healy J., Saul N., Großberger L., 2018, @doi [Journal of Open Source Software] 10.21105/joss.00861 , 3, 861

    2018 doi

  36. [44]

    Mereghetti S., et al., 2020, @doi [The Astrophysical Journal Letters] 10.3847/2041-8213/aba2cf , 898, L29

    2020 doi

  37. [45]

    Michilli D., et al., 2021, @doi [The Astrophysical Journal] 10.3847/1538-4357/abe626 , 910, 147

    2021 doi

  38. [46]

    NAN R., et al., 2011, @doi [International Journal of Modern Physics D] 10.1142/s0218271811019335 , 20, 989–1024

    2011 doi

  39. [47]

    Nimmo K., et al., 2021, Nature Astronomy, 5, 594

    2021

  40. [48]

    Nimmo K., et al., 2022, @doi [Nature Astronomy] 10.1038/s41550-021-01569-9 , https://ui.adsabs.harvard.edu/abs/2022NatAs...6..393N 6, 393

    2022 doi

  41. [49]

    Pandhi A., et al., 2024, @doi [ ] 10.3847/1538-4357/ad40aa , https://ui.adsabs.harvard.edu/abs/2024ApJ...968...50P 968, 50

    2024 doi

  42. [50]

    B., Majid W

    Pearlman A. B., Majid W. A., Prince T. A., Nimmo K., Hessels J. W. T., Naudet C. J., Kocz J., 2020, @doi [ ] 10.3847/2041-8213/abca31 , https://ui.adsabs.harvard.edu/abs/2020ApJ...905L..27P 905, L27

    2020 doi

  43. [51]

    B., et al., 2024, @doi [Nature Astronomy] 10.1038/s41550-024-02386-6

    Pearlman A. B., et al., 2024, @doi [Nature Astronomy] 10.1038/s41550-024-02386-6

    2024 doi

  44. [52]

    Petroff E., Hessels J., Lorimer D., 2019, @doi [The Astronomy and Astrophysics Review] 10.1007/s00159-019-0116-6 , 27

    2019 doi

  45. [53]

    Petroff E., Hessels J. W. T., Lorimer D. R., 2022, @doi [ ] 10.1007/s00159-022-00139-w , https://ui.adsabs.harvard.edu/abs/2022A&ARv..30....2P 30, 2

    2022 doi

  46. [54]

    Pleunis Z., et al., 2021, @doi [ ] 10.3847/1538-4357/ac33ac , https://ui.adsabs.harvard.edu/abs/2021ApJ...923....1P 923, 1

    2021 doi

  47. [55]

    Springer Berlin Heidelberg, Berlin, Heidelberg, pp 55--69, @doi 10.1007/3-540-49430-8_3 , https://doi.org/10.1007/3-540-49430-8_3

    Prechelt L., 1998, Early Stopping - But When?. Springer Berlin Heidelberg, Berlin, Heidelberg, pp 55--69, @doi 10.1007/3-540-49430-8_3 , https://doi.org/10.1007/3-540-49430-8_3

    1998 doi

  48. [56]

    C., et al., 2019, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stz958 , 486, 3636–3646

    Price D. C., et al., 2019, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stz958 , 486, 3636–3646

    2019 doi

  49. [57]

    Ravi V., Loeb A., 2019, @doi [ ] 10.3847/1538-4357/ab0748 , https://ui.adsabs.harvard.edu/abs/2019ApJ...874...72R 874, 72

    2019 doi

  50. [58]

    Ravi V., et al., 2016, @doi [Science] 10.1126/science.aaf6807 , https://ui.adsabs.harvard.edu/abs/2016Sci...354.1249R 354, 1249

    2016 doi

  51. [59]

    arXiv:2103.11251

    Rudin C., Chen C., Chen Z., Huang H., Semenova L., Zhong C., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2103.11251 , https://ui.adsabs.harvard.edu/abs/2021arXiv210311251R p. arXiv:2103.11251

    2021 doi arXiv

  52. [60]

    M., 2024, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stae1972 , 533, 3283–3295

    Sharma A., Rajpaul V. M., 2024, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stae1972 , 533, 3283–3295

    2024 doi

  53. [61]

    Sherman M., Connor L., Ravi V., Law C., DSA-2000 Collaboration 2024, in American Astronomical Society Meeting Abstracts. p. 261.04

    2000

  54. [62]

    arXiv:2410.07307

    Shin K., et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2410.07307 , https://ui.adsabs.harvard.edu/abs/2024arXiv241007307S p. arXiv:2410.07307

    2024 doi arXiv

  55. [63]

    P., et al., 2023, @doi [Nature Astronomy] 10.1038/s41550-023-02101-x , https://ui.adsabs.harvard.edu/abs/2023NatAs...7.1486S 7, 1486

    Snelders M. P., et al., 2023, @doi [Nature Astronomy] 10.1038/s41550-023-02101-x , https://ui.adsabs.harvard.edu/abs/2023NatAs...7.1486S 7, 1486

    2023 doi

  56. [64]

    G., et al., 2016, @doi [ ] 10.1038/nature17168 , https://ui.adsabs.harvard.edu/abs/2016Natur.531..202S 531, 202

    Spitler L. G., et al., 2016, @doi [ ] 10.1038/nature17168 , https://ui.adsabs.harvard.edu/abs/2016Natur.531..202S 531, 202

    2016 doi

  57. [65]

    pp 1--4, @doi 10.1109/IWSSIP.2009.5367700

    State L., Constantin D., Sararu C., 2009, in 2009 16th International Conference on Systems, Signals and Image Processing. pp 1--4, @doi 10.1109/IWSSIP.2009.5367700

    2009

  58. [66]

    Sun W.-P., Zhang J.-G., Li Y., Hou W.-T., Zhang F.-W., Zhang J.-F., Zhang X., 2024, Exploring the Key Features of Repeating Fast Radio Bursts with Machine Learning ( @eprint arXiv 2409.11173 ), https://arxiv.org/abs/2409.11173

    2024 arXiv

  59. [67]

    C., Sheridan R

    Svetnik V., Liaw A., Tong C., Culberson J. C., Sheridan R. P., Feuston B. P., 2003, Journal of chemical information and computer sciences, 43 6, 1947

    2003

  60. [68]

    Thornton D., et al., 2013, @doi [Science] 10.1126/science.1236789 , https://ui.adsabs.harvard.edu/abs/2013Sci...341...53T 341, 53

    2013 doi

  61. [69]

    Vanderlinde K., et al., 2019, in Canadian Long Range Plan for Astronomy and Astrophysics White Papers. p. 28 ( @eprint arXiv 1911.01777 ), @doi 10.5281/zenodo.3765414

    2019 arXiv

  62. [70]

    Wang Z., Bovik A., Sheikh H., Simoncelli E., 2004, @doi [IEEE Transactions on Image Processing] 10.1109/TIP.2003.819861 , 13, 600

    2004

  63. [71]

    Wold S., Esbensen K., Geladi P., 1987, @doi [Chemometrics and Intelligent Laboratory Systems] https://doi.org/10.1016/0169-7439(87)80084-9 , 2, 37

    1987 doi

  64. [72]

    L., 2015, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/15/10/002 , https://ui.adsabs.harvard.edu/abs/2015RAA....15.1629X 15, 1629

    Xu J., Han J. L., 2015, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/15/10/002 , https://ui.adsabs.harvard.edu/abs/2015RAA....15.1629X 15, 1629

    2015 doi

  65. [73]

    Ying X., 2019, in Journal of Physics Conference Series. IOP, p. 022022, @doi 10.1088/1742-6596/1168/2/022022

    2019 doi

  66. [74]

    Zhang Y.-K., et al., 2022, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/ac98f7 , https://ui.adsabs.harvard.edu/abs/2022RAA....22l4002Z 22, 124002

    2022 doi

  67. [75]

    Zhou B., Li X., Wang T., Fan Y.-Z., Wei D.-M., 2014, @doi [Phys. Rev. D] 10.1103/PhysRevD.89.107303 , 89, 107303

    2014 doi

  68. [76]

    Zhu-Ge J.-M., Luo J.-W., Zhang B., 2023, @doi [ ] 10.1093/mnras/stac3599 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.519.1823Z 519, 1823

    2023 doi

  69. [77]

    van der Maaten L., Hinton G., 2008, Journal of Machine Learning Research, 9, 2579

    2008

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.