Pith. sign in

REVIEW 4 major objections 5 minor 65 references

A convolutional autoencoder separates fast radio bursts into three distinct morphological classes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 06:03 UTC pith:DGUQAWDZ

load-bearing objection A plausible but not fully data-driven FRB morphology paper: the scattered/unscattered broadband split is new and worth testing, but the simulator's hand-set priors do much of the work. the 4 major comments →

arxiv 2607.13148 v1 pith:DGUQAWDZ submitted 2026-07-14 astro-ph.HE astro-ph.CO

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

Bo Lin Fan , Ren\'ee Hlo\v{z}ek , Antonio Herrera Martin This is my paper
classification astro-ph.HE astro-ph.CO
keywords fast radio burstsconvolutional autoencoderunsupervised classificationmorphologywaterfall plotsrepeatabilityscatteringCHIME/FRB
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that an unsupervised convolutional autoencoder trained on simulated burst images can map real CHIME/FRB waterfall plots into a compact latent space where three morphological subgroups stand out from the continuous majority: narrowband downward-drifting bursts, and two varieties of simple broadband bursts distinguished by scattering. The same latent space supports a repeatability classifier that recovers 86% of repeater bursts, comparable to parameter-based methods, even though repeaters and one-off bursts overlap substantially in morphology. If correct, this provides a reproducible, data-driven way to define FRB morphologies, independently recovering at least some previously visual classes and revealing a scattering-based split that manual inspection had not cleanly separated.

Core claim

The paper's central claim is that an unsupervised convolutional autoencoder, trained on synthetic bursts derived from DR1 fitburst statistics, learns a latent space of CHIME/FRB waterfalls in which three morphological groups separate from the continuous bulk of the population: G7 (narrowband, long-duration bursts with downward-drifting sub-bursts), G4 (temporally short, full-bandwidth bursts with strong scattering), and G6 (equally broadband but minimally scattered). These groups persist when the same network is applied to the much larger DR2 sample, while intermediate DR1 clusters merge into a continuum. A classifier head on the same latent space identifies 86% of true repeating bursts (rec

What carries the argument

The central object is the 256-dimensional latent representation produced by a convolutional autoencoder trained on 7,500 synthetic waterfall plots. To make those simulations, the authors fit kernel-density estimates to the DR1 fitburst parameters of one-off and repeating bursts separately, draw synthetic burst parameters from them, add one to four sub-bursts with uniform temporal and frequency offsets, then inject CHIME-like RFI masks and Gaussian noise. Real DR1 and DR2 waterfalls are encoded through the same network, and the latent codes are projected by PCA plus UMAP and clustered with HDBSCAN; cluster membership defines morphology, while a supervised classification head attached to the l

Load-bearing premise

The whole classification inherits the assumption that the synthetic bursts, drawn from DR1 parameter distributions plus hand-picked sub-burst offsets, resemble real CHIME/FRB waterfalls closely enough that features learned on simulations transfer to real data; if the simulator's sub-burst or scattering prescriptions are unrealistic, the discovered clusters could be artifacts of the simulator rather than true population structure.

What would settle it

Train the identical autoencoder on a simulation set with sub-burst offsets scrambled (for example, all upward drift or no drift at all) and re-cluster real DR2 waterfalls: if G7 still emerges, the drifting-narrowband class is not being imposed by the training prior, but if it disappears, the class is a simulator artifact. Alternatively, apply the same pipeline to CHIME baseband data with sub-millisecond resolution: if G7-like downward drift is resolved there, it is intrinsic, not an artifact of the 0.983 ms intensity sampling.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Three morphological groups (G7 narrowband drifting, G4 broadband scattered, G6 broadband unscattered) are separable from the continuous bulk of FRB morphology in both DR1 and DR2.
  • G4 and G6 likely reflect a single simple broadband burst population, differing only by whether a plasma screen scatters the signal before it reaches the observer.
  • Morphology alone cannot classify repeatability: repeaters and one-off bursts overlap heavily, and some repeating sources emit bursts that look like typical one-offs.
  • A latent-space classifier catches 86% of true repeaters at precision around 0.4 in DR1 and 0.35 in DR2, comparable to parameter-based repeatability classifiers.
  • The downward linear drifting morphology previously defined visually is recovered without labels, and the scattering split between G4 and G6 is a distinction that visual classification did not cleanly make.

Where Pith is reading between the lines

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

  • A natural next test is to run the same simulator-to-latent pipeline on CHIME baseband data: if G7 persists at high time resolution, it is intrinsic; if it weakens or vanishes, instrumental beam or sidelobe effects are responsible for the narrowband appearance.
  • The G4/G6 scattering split, if real, makes scattering time a probe of the line-of-sight environment, suggesting that clean one-off bursts (G6) trace relatively under-dense propagation paths while G4 bursts pass through denser plasma screens.
  • Because training used only DR1 distributions yet classifications transfer to DR2, the latent space appears stable across catalogue versions; applying the same approach to other telescopes or larger samples would test whether the three classes are universal.
  • The simulator's sub-burst drift model is the main prior in the pipeline; systematically varying the drift direction and offset distributions would reveal how much of G7 is learned from the data rather than imposed by the training set.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper trains a convolutional autoencoder (CAE) with a supervised classification head on synthetic CHIME/FRB waterfall plots generated from KDEs of the DR1 fitburst parameter distributions plus hand-specified priors for sub-burst structure, scattering, and noise. The 256-dimensional latent space is reduced with PCA/UMAP and clustered with HDBSCAN, applied to both DR1 and DR2. The authors report three robust morphological classes: narrowband downward-drifting bursts (G7), and two broader classes separated by scattering (G4 vs G6); they also report a repeatability classifier with recall ~0.85–0.86 and precision ~0.40 (DR1) / ~0.35 (DR2). They argue that most bursts lie on a continuous morphology distribution, with only these extremes forming distinct groups.

Significance. If the clusters are robust, the paper provides a genuinely useful data-driven morphological taxonomy for the largest available CHIME/FRB sample, and independently recovering the known linear-drift repeater morphology from waterfall images is a concrete, nontrivial success. The use of DR1-based simulations for the DR2 analysis partially mitigates circularity. However, the central assertion that G4/G6/G7 are data-driven classes rests on an unvalidated simulator; the hand-chosen priors for drift and scattering shape exactly the features that define those clusters. The reported inconsistencies in HDBSCAN min_cluster_size and in the DR1 confusion matrix also need to be resolved before the quantitative claims are fully reproducible.

major comments (4)
  1. [§2.2] The feature extractor is trained exclusively on simulated waterfalls. Sub-burst time offsets (5–10 ms), frequency offsets (−200 to +10 MHz), sub-burst count (1–4), and scattering index (−4.4 to −3.5) are hand-chosen, not calibrated to observed DR1/DR2 sub-burst or scattering statistics. Because G7 is defined by downward-drifting sub-bursts and G4 by strong scattering, the clustering is directly shaped by these priors. The paper gives no ablation varying the priors and no quantitative comparison of simulated vs real waterfall statistics. The DR2 analysis is partly independent, but the same unvalidated encoder is applied to DR2. The claim that G4/G6/G7 are data-driven classes is therefore not yet secured. Please add prior/robustness tests, or validate the simulator against real data statistics.
  2. [§3.4 and Table 2] The text sets HDBSCAN min_cluster_size = 10, while Table 2 lists 20. This hyperparameter directly controls which clusters are resolved, so the discrepancy affects all cluster assignments and counts in Figures 4 and 5. State the exact value used, fix the inconsistency, and if both settings were tried, report the effect on cluster stability and membership.
  3. [Figure 6A] The DR1 confusion matrix is inconsistent with the catalogue size. §2.1 gives 536 DR1 bursts (98 repeaters, 438 one-offs), but the matrix contains entries summing >1800 (398, 1349, 75). The quoted recall (0.85) and precision (0.40) imply roughly 83 true positives, 15 false negatives, and ~125 false positives — none of the printed numbers match. This must be corrected and the DR1 repeatability metrics recomputed/re-reported.
  4. [§2.2] The KDEs used to generate the training set are a core component, but no kernel choice or bandwidths are reported, only parameter ranges (Table 1). Without these values, the simulation, and therefore the learned embeddings, are not reproducible. Provide bandwidths (or code) and describe the KDE construction in enough detail to allow replication.
minor comments (5)
  1. [Abstract/§4.3.5] Repeated phrase 'visually inspection' should be 'visual inspection'; there are other minor grammatical issues throughout that should be cleaned up.
  2. [§2.2] The sentence 'we simulate 2500 one-off pulses for both the one-off and repeater groups' is ambiguous. Clarify the sample sizes for the three simulation subsets: single one-off, single repeater, and multi-sub-burst events.
  3. [Equation (3)] Equation (3) introduces n ∝ r^k with no definition of n or r, and it is not used elsewhere in the analysis. Either substantiate the physical model or remove the equation.
  4. [Figures 6 and 7] Panels B–D of Figures 6 and 7 lack clear legends for 'Predicted One-Off', 'Predicted Repeater', and 'True Repeater'. Define the colour scheme in each caption, and ensure axis labels are visible in all panels.
  5. [Table 1] Table 1 is formatted inconsistently: the row for 'ref freq' reports a dash, and some ranges appear in descending order (e.g., [0.126,0.001]) without comment. Standardize the table and clarify the parameter ranges.

Circularity Check

1 steps flagged

No significant circularity: the central claims are grounded by out-of-sample DR2 analysis and external benchmarks; one mild case where a morphology hand-seeded into the training simulator is presented as independently recovered.

specific steps
  1. renaming known result [Abstract; §2.2 (Simulating FRB Observations using fitburst)]
    ""Central frequency differences of each sub-burst and relative to the preceding burst were sampled from a random uniform distribution between -200 MHz and 10 MHz. This is to encompass both typical downward-drifting bursts and the rarer upward-drifting events reported in the literature (Sand et al. 2025; Faber et al. 2024)." ... "We find that our approach is able to independently recover the downward linear drifting burst morphologies previously defined through visually inspection.""

    The CAE training set is deliberately seeded with the very morphologies later reported as discovered: downward-drifting sub-bursts (uniform frequency offsets -200 to +10 MHz relative to the preceding burst; §2.2) and scattered bursts (scattering index drawn uniformly in [-4.4,-3.5]). Those hand-chosen priors build drift- and scattering-sensitive axes into the encoder a priori, so the latent clusters G7 (drifting, narrowband) and G4 (scattered, broadband) re-express the simulation's own inputs under new cluster labels. The abstract's 'independently recover' is therefore overstated: the drifting morphology entered via the training set, not as an unprompted discovery. This is mild, because the clustering itself is unsupervised on real waterfall data and the DR2 persistence is an out-of-sample

full rationale

The derivation chain: DR1 fitburst parameters → KDEs → simulated waterfalls (with hand-chosen sub-burst/scattering priors) → CAE latent space → PCA/UMAP/HDBSCAN clusters on real DR1/DR2 embeddings, plus a supervised classifier head trained on those simulations for repeatability. No step reduces by an equation to a fitted value. The repeatability metrics (Precision 0.40/0.35, Recall 0.85/0.86; Eq. 2) are evaluated against real catalogue labels via confusion matrices on the real data (Figures 6-7) and are benchmarked externally (Kharel et al. 2025 recall ~0.85; Sun et al. 2026a,b precision ~0.4), so the 86% recall is not a fitted input renamed as a prediction. DR2 is a genuine out-of-sample test: the KDEs were fit to DR1 only, and the persistent G4/G6/G7 clusters appear in DR2 embeddings, so the central claim retains independent content. The cited empirical anchors (repeaters narrow-band/long-duration; Pleunis et al. 2021) are externally reproduced by independent groups (Chen et al. 2021; Sharma & Rajpaul 2024; Sun et al. 2026a,b); even if the large CHIME collaboration author lists overlap with the present authors, no load-bearing claim rests on a self-citation, and no uniqueness theorem or ansatz is imported from the authors' own prior work. The one mild circular step is flagged above: §2.2 deliberately seeds drifting/scattered templates into the training set, and the abstract frames recovery of that same morphology as 'independent.' The paper itself partially mitigates this by acknowledging the correspondence and by testing on DR2; the scattered/unscattered split (G4 vs G6) is the genuinely new content. Per the review rule, I also weighed the manuscript's own limitation passages: §2.2 states 'the distribution of scattering indices is not known for the CHIME/FRB DR1 sample' and §3.4 says 'further optimization of our clustering procedure is left for future work' — these are honest robustness notes, not admissions of circularity. One non-circular reproducibility inconsistency exists: §3.4 states min_cluster_size = 10 while Table 2 lists min cluster size = 20. Overall, the derivation is self-contained against external benchmarks and out-of-sample tests, so the appropriate score is 2, reflecting the single mild seeded-morphology re-labeling rather than any construction-level reduction.

Axiom & Free-Parameter Ledger

10 free parameters · 6 axioms · 0 invented entities

The central claim rests on a simulation-based transfer-learning pipeline. The simulations are built from DR1 fitburst parameters with several hand-chosen ranges for sub-burst structure and scattering. The main free parameters are the KDE bandwidths, sub-burst offsets, scattering index range, central frequency range, and clustering hyperparameters. No new physical entities are introduced; the paper's interpretation uses existing models (e.g., Metzger et al. 2022) only as discussion.

free parameters (10)
  • KDE bandwidths for DR1 fitburst parameter distributions
    Kernel density estimates over temporal width, spectral index, spectral running, and scattering timescale require bandwidth selection, not stated (Section 2.2).
  • sub-burst time offset range 5-10 ms
    Chosen by hand in Section 2.2 to simulate multi-component bursts.
  • sub-burst central frequency offset uniform [-200, +10] MHz
    Hand-chosen range to encompass downward- and upward-drifting bursts (Section 2.2).
  • sub-burst count 1-4 uniform
    Choice based on Faber et al. 2024, but not directly fitted to the data (Section 2.2).
  • scattering index uniform [-4.4, -3.5]
    Sampled from literature range (Petroff et al. 2016), not fit to FRB data (Section 2.2).
  • central frequency uniform [200, 1200] MHz
    Used because true central frequencies may lie outside band; range chosen by hand (Section 2.2).
  • CAE latent dimension 256
    Architectural choice (Section 3.1).
  • UMAP n_neighbors=5, min_dist=0.01
    Hyperparameters chosen for visualization; not optimized (Table 2).
  • HDBSCAN min_cluster_size (10 vs 20)
    Inconsistent between Section 3.4 (10) and Table 2 (20).
  • Gaussian smoothing kernel 5x5 and tophat 100x1
    Preprocessing choices (Section 2.3).
axioms (6)
  • domain assumption Simulated fitburst waterfalls with added noise are representative of real CHIME/FRB total-intensity waterfalls
    Section 2.2 assumes the simulation pipeline captures the diversity of observed FRB properties.
  • domain assumption The KDEs fitted to DR1 one-off and repeater parameter distributions capture the true population distributions of both classes
    Section 2.2 constructs KDEs from DR1 and assumes they generalize.
  • domain assumption Labels of simulated bursts as repeater/one-off, based on which DR1 distribution they are drawn from, correspond to true physical repeatability
    Section 3.2 uses these labels to train the classification head.
  • domain assumption The CAE features learned on simulations transfer to real CHIME/FRB data
    Section 4 applies the simulation-trained model to real DR1 and DR2 data.
  • domain assumption fitburst model parameters accurately describe FRB morphology
    Section 2.1 relies on fitburst-derived parameters to build simulations and interpret clusters.
  • domain assumption The CHIME/FRB RFI masking and noise model used in simulations matches the real data
    Section 2.2 uses DR1 RFI masking probabilities and RMS noise distributions.

pith-pipeline@v1.3.0-alltime-deepseek · 20637 in / 12842 out tokens · 112088 ms · 2026-08-02T06:03:04.329547+00:00 · methodology

0 comments
read the original abstract

Understanding the morphology of fast radio bursts (FRB), and whether all sources repeat, are key challenges that are becoming more tractable given the increase in data from surveys such as the Canadian Hydrogen Intensity Mapping Experiment FRB project (CHIME/FRB). We present a Convolutional Autoencoder unsupervised classifier for separating the CHIME/FRB data into morphological classes. This data-driven approach is more reproducible than visual inspection, since groupings are learned from the data itself and not subject to differences between expert annotations. While most bursts occupy a similar area of morphological parameter space, we identify three classes of bursts separate from the general FRB population. While one class contains bursts with short bandwidth, and downward-drifting sub-burst structure, the characteristic bursts of other classes have very short temporal width, and occupy the entire CHIME observing band. We identify two distinct subgroups of temporally short, simple broadband bursts; one with minimal scattering and the other with higher scattering. As an additional output of our classifier, we provide a binary FRB repeatability classification, and train the classifications on simulations that mimic the first FRB catalogue from CHIME/FRB. We are able to correctly identify 86$\%$ of repeater bursts. We find that our approach is able to independently recover the downward linear drifting burst morphologies previously defined through visually inspection. Overall, we find that although there exists FRB subgroups with higher or lower proportion of repeaters, there is substantial overlap between the morphological properties of repeaters and one-off bursts consistent with previous studies.

Figures

Figures reproduced from arXiv: 2607.13148 by Antonio Herrera Martin, Bo Lin Fan, Ren\'ee Hlo\v{z}ek.

Figure 1
Figure 1. Figure 1: Examples of bursts modelled using the fit￾burst implementation in both the second (middle panel) and first (right panel) CHIME/FRB data release papers, com￾pared to the CHIME/FRB total intensity image (left panel). The top row shows the data and model for FRB 20190219B, which shows some changes to the model parameters between data releases, while the bottom row is for FRB 20181213A, a burst that is consist… view at source ↗
Figure 2
Figure 2. Figure 2: The flow diagram of the semi-supervised CAE used to determine FRB morphology while simultaneously classifying for FRB repeatability. A waterfall plot of a given burst (left) is encoded into a lower dimensional representation, the latent space. The model performance is then evaluated as a sum of its ability to reconstruct the original burst waterfall plot from this latent representation, and its ability to … view at source ↗
Figure 3
Figure 3. Figure 3: In order to simulate bursts to train our CAE, we first sample parameters shown in the box A) from KDE fitted to repeating or non-repeating FRBs in the first CHIME/FRB dataset. These parameters are then used to simulated burst waterfall plots using fitburst (B). Finally, Gaussian noise is applied and certain frequency bands are masked to simulate CHIME instrumental noise and RFI respectively (C), and the pl… view at source ↗
Figure 4
Figure 4. Figure 4: Unsupervised clustering of the CHIME/FRB DR1 dataset Top panel: the UMAP diagram showing the clustering into seven subgroups based on the latent parameters of the CAE model trained on simulations, applied to the DR1 data set. Some clusters (G4, G6, and G7) form distinct, well-separated groups corresponding to previously identified morphologies, while intermediate clusters (G1, G2, G3, G5) show more gradual… view at source ↗
Figure 5
Figure 5. Figure 5: Unsupervised clustering of the CHIME/FRB DR2 dataset. Left panel: the UMAP diagram showing the clustering into seven subgroups based on the latent parameters of the CAE model trained on simulations based on DR1 parameter distributions, applied to the DR2 data set. Clustering of the CHIME/FRB DR2 dataset further enforces our finding that while the majority of FRBs overlap in parameter space, there are three… view at source ↗
Figure 6
Figure 6. Figure 6: Top panel: Confusion matrices from the CAE classification applied to the CHIME/FRB DR1 data set. 20% of the true repeaters where classified as one-off bursts, and 16% of the true one-off bursts were misclassified as re￾peaters. Bottom panel: The UMAP representations of the latent CAE parameters does not show obvious clustering into two distinct groups according to predicted repeatability. Due to the small … view at source ↗
Figure 7
Figure 7. Figure 7: A) Confusion matrices from the CAE classification applied to the CHIME/FRB DR2 data set. 15% of the repeaters are misclassified as one-off bursts, while the fraction of one-off bursts misclassified as repeaters grew to 45% in DR2. The UMAP representations of the latent CAE parameters shown in C) highlights that the latent parameter distribution of predicted repeating and predicted single/one-off bursts ove… view at source ↗
Figure 8
Figure 8. Figure 8: Parameter distributions from the fitburst model for the morphological groupings described in Section 4.1 obtained when applying our CAE unsupervised clustering to the CHIME/FRB DR1 as compared to the full one-off and repeater dis￾tributions of CHIME/FRB DR1 and DR2. Spectral running, temporal width, and frequency span are shown as these have been established to be most distinguishing between repeater and o… view at source ↗
Figure 9
Figure 9. Figure 9: Examples of repeating bursts in DR2 found in G4 and G6, clusters comprised predominantly of one-off bursts. An additional burst from these repeating sources not consid￾ered part of these primarily one-off sources are also shown to the left, and displays morphology more typical of repeaters. Despite being repeaters, these bursts show broadband and, in the case of G4, negative power-law drifting burst struct… view at source ↗
Figure 10
Figure 10. Figure 10: Individual bursts observed from the repeating burst FRB 20190905A. Bursts are numbered in order of arrival time, with 1 being the oldest recorded burst. Predictions from our classification pipeline are indicated by the colour of the point, with orange denoting that it was predicted to be a repeater, while blue denotes that the observation was predicted to be a one-off burst from its morphological clusteri… view at source ↗
Figure 11
Figure 11. Figure 11: The full suite of parameter distributions from the fitburst model for the morphological groupings described in Section 4.1 obtained when applying our CAE unsupervised clustering to the CHIME/FRB DR1. A subset of parameter distributions relevant to repeatability are shown in [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Parameter distributions from the fitburst model for the morphological groupings described in Section 4.3 obtained when applying our CAE unsupervised clustering to the CHIME/FRB DR2. A subset of parameter distributions relevant to repeatability are shown in [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

65 extracted references · 8 canonical work pages · 2 internal anchors

  1. [1]

    arXiv e-prints , keywords =

    Discovery of 30 Repeating Fast Radio Burst Sources and Uniform Population Statistics of 80 Repeating Sources from CHIME/FRB. arXiv e-prints , keywords =. doi:10.48550/arXiv.2605.08410 , archivePrefix =. 2605.08410 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2605.08410

  2. [2]

    arXiv e-prints , keywords =

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv e-prints , keywords =. doi:10.48550/arXiv.1802.03426 , archivePrefix =. 1802.03426 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1802.03426

  3. [3]

    arXiv e-prints , keywords =

    A Tutorial on Principal Component Analysis. arXiv e-prints , keywords =. doi:10.48550/arXiv.1404.1100 , archivePrefix =. 1404.1100 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1404.1100

  4. [4]

    G. E. Hinton and R. R. Salakhutdinov , title =. Science , volume =. 2006 , doi =

    2006

  5. [5]

    A density-based algorithm for discovering clusters in large spatial databases with noise , year =

    Ester, Martin and Kriegel, Hans-Peter and Sander, J\". A density-based algorithm for discovering clusters in large spatial databases with noise , year =. Proceedings of the Second International Conference on Knowledge Discovery and Data Mining , pages =

  6. [6]

    Science , keywords =

    A Bright Millisecond Radio Burst of Extragalactic Origin. Science , keywords =. doi:10.1126/science.1147532 , archivePrefix =. 0709.4301 , primaryClass =

    Pith/arXiv arXiv

  7. [7]

    , keywords =

    FRB 200428: An Impact between an Asteroid and a Magnetar. , keywords =. doi:10.3847/2041-8213/aba83c , archivePrefix =. 2006.04601 , primaryClass =

    Pith/arXiv arXiv 2041

  8. [8]

    Wang, F. Y. and Zhang, G. Q. and Dai, Z. G. and Cheng, K. S. , Da =. Repeating fast radio burst 20201124A originates from a magnetar/Be star binary , Ty =. Nature Communications , Number =. 2022 , Bdsk-Url-1 =. doi:10.1038/s41467-022-31923-y , Id =

    doi:10.1038/s41467-022-31923-y 2022

  9. [9]

    , keywords =

    The Rarity of Repeating Fast Radio Bursts from Binary Neutron Star Mergers. , keywords =. doi:10.3847/1538-4357/ab7c5c , archivePrefix =. 2003.01919 , primaryClass =

    Pith/arXiv arXiv 2003

  10. [10]

    International Journal of Modern Physics D , keywords =

    The Five-Hundred Aperture Spherical Radio Telescope (fast) Project. International Journal of Modern Physics D , keywords =. doi:10.1142/S0218271811019335 , archivePrefix =. 1105.3794 , primaryClass =

    Pith/arXiv arXiv

  11. [11]

    , keywords =

    Morphologies of Bright Complex Fast Radio Bursts with CHIME/FRB Voltage Data. , keywords =. doi:10.3847/1538-4357/ad59aa , archivePrefix =. 2312.14133 , primaryClass =

    Pith/arXiv arXiv

  12. [12]

    , keywords =

    Radio Bursts with Extragalactic Spectral Characteristics Show Terrestrial Origins. , keywords =. doi:10.1088/0004-637X/727/1/18 , archivePrefix =. 1009.5392 , primaryClass =

    Pith/arXiv arXiv

  13. [13]

    Frabjous: Deep Learning Fast Radio Burst Morphologies

    Frabjous: Deep Learning Fast Radio Burst Morphologies. arXiv e-prints , keywords =. doi:10.48550/arXiv.2507.14854 , archivePrefix =. 2507.14854 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2507.14854

  14. [14]

    , keywords =

    An Overview of CHIME, the Canadian Hydrogen Intensity Mapping Experiment. , keywords =. doi:10.3847/1538-4365/ac6fd9 , archivePrefix =. 2201.07869 , primaryClass =

    Pith/arXiv arXiv

  15. [15]

    , keywords =

    Morphology of 35 Repeating Fast Radio Burst Sources at Microsecond Time Scales with CHIME/FRB. , keywords =. doi:10.3847/1538-4357/adf844 , archivePrefix =. 2411.02870 , primaryClass =

    Pith/arXiv arXiv

  16. [16]

    , keywords =

    The CHIME Fast Radio Burst Project: System Overview. , keywords =. doi:10.3847/1538-4357/aad188 , archivePrefix =. 1803.11235 , primaryClass =

    Pith/arXiv arXiv

  17. [17]

    , keywords =

    Updating the First CHIME/FRB Catalog of Fast Radio Bursts with Baseband Data. , keywords =. doi:10.3847/1538-4357/ad464b , archivePrefix =. 2311.00111 , primaryClass =

    Pith/arXiv arXiv

  18. [18]

    , keywords =

    Five new fast radio bursts from the HTRU high-latitude survey at Parkes: first evidence for two-component bursts. , keywords =. doi:10.1093/mnrasl/slw069 , archivePrefix =. 1511.07746 , primaryClass =

    Pith/arXiv arXiv

  19. [19]

    , keywords =

    Representation learning for fast radio burst dynamic spectra. , keywords =. doi:10.1093/mnras/staf306 , archivePrefix =. 2412.12394 , primaryClass =

    Pith/arXiv arXiv

  20. [20]

    arXiv e-prints , keywords =

    Deep Learning using Rectified Linear Units (ReLU). arXiv e-prints , keywords =. doi:10.48550/arXiv.1803.08375 , archivePrefix =. 1803.08375 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1803.08375

  21. [21]

    arXiv e-prints , keywords =

    Adam: A Method for Stochastic Optimization. arXiv e-prints , keywords =. doi:10.48550/arXiv.1412.6980 , archivePrefix =. 1412.6980 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980

  22. [22]

    Non-Repeating FRBs: A Deep Learning Approach To Morphological Characterization

    Repeating vs. Non-Repeating FRBs: A Deep Learning Approach To Morphological Characterization. arXiv e-prints , keywords =. doi:10.48550/arXiv.2509.06208 , archivePrefix =. 2509.06208 , primaryClass =

    doi:10.48550/arxiv.2509.06208

  23. [23]

    , keywords =

    Fast Radio Burst Morphology in the First CHIME/FRB Catalog. , keywords =. doi:10.3847/1538-4357/ac33ac , archivePrefix =. 2106.04356 , primaryClass =

    Pith/arXiv arXiv

  24. [24]

    , keywords =

    Flux Calibration of CHIME/FRB Intensity Data. , keywords =. doi:10.3847/1538-3881/acec78 , archivePrefix =. 2305.11302 , primaryClass =

    Pith/arXiv arXiv

  25. [25]

    , keywords =

    A probe of the maximum energetics of fast radio bursts through a prolific repeating source. , keywords =. doi:10.1093/mnras/staf1937 , archivePrefix =. 2410.17024 , primaryClass =

    arXiv

  26. [26]

    Nature Astronomy , keywords =

    A link between repeating and non-repeating fast radio bursts through their energy distributions. Nature Astronomy , keywords =. doi:10.1038/s41550-023-02153-z , archivePrefix =. 2306.15505 , primaryClass =

    arXiv

  27. [27]

    , keywords =

    A Steadily Declining Dispersion Measure for the Repeating Fast Radio Burst FRB 20220529A: Evidence for a Fast Radio Burst Engine Embedded in an Expanding Supernova Remnant. , keywords =. doi:10.3847/2041-8213/ae52f8 , archivePrefix =. 2602.22309 , primaryClass =

    arXiv 2041

  28. [28]

    arXiv e-prints , keywords =

    The Real and Pseudo Dispersion Measures of FRB -0.5ex 20220912A. arXiv e-prints , keywords =. doi:10.48550/arXiv.2604.01825 , archivePrefix =. 2604.01825 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2604.01825

  29. [29]

    , keywords =

    Magnetospheric origin of a fast radio burst constrained using scintillation. , keywords =. doi:10.1038/s41586-024-08297-w , archivePrefix =. 2406.11053 , primaryClass =

    Pith/arXiv arXiv

  30. [30]

    arXiv e-prints , keywords =

    FRB scattering statistics through the CGM are sensitive to morphology and intermittency. arXiv e-prints , keywords =. doi:10.48550/arXiv.2603.00336 , archivePrefix =. 2603.00336 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2603.00336

  31. [31]

    , keywords =

    Scintillometry of fast radio bursts: Resolution effects in two-screen models. , keywords =. doi:10.1051/0004-6361/202554202 , archivePrefix =. 2505.04576 , primaryClass =

    Pith/arXiv arXiv

  32. [32]

    arXiv e-prints , keywords =

    The CHIME/FRB Discovery of the Extremely Active Fast Radio Burst Source FRB 20240114A. arXiv e-prints , keywords =. doi:10.48550/arXiv.2505.13297 , archivePrefix =. 2505.13297 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2505.13297

  33. [33]

    , keywords =

    Deep Synoptic Array Science: Searching for Long Duration Radio Transients with the DSA-110. , keywords =. doi:10.1088/1538-3873/ae390e , archivePrefix =. 2510.18136 , primaryClass =

    arXiv

  34. [34]

    Monthly Notices of the Royal Astronomical Society , volume =

    Rajwade, K M and Bezuidenhout, M C and Caleb, M and Driessen, L N and Jankowski, F and Malenta, M and Morello, V and Sanidas, S and Stappers, B W and Surnis, M P and Barr, E D and Chen, W and Kramer, M and Wu, J and Buchner, S and Serylak, M and Combes, F and Fong, W and Gupta, N and Jagannathan, P and Kilpatrick, C D and Krogager, J-K and Noterdaeme, P a...

    2022

  35. [35]

    , keywords =

    The commensal real-time ASKAP fast transient incoherent-sum survey. , keywords =. doi:10.1017/pasa.2025.8 , archivePrefix =. 2408.02083 , primaryClass =

    Pith/arXiv arXiv 2025

  36. [36]

    , keywords =

    Are Persistent Emission Luminosity and Rotation Measure of Fast Radio Bursts Related?. , keywords =. doi:10.3847/1538-4357/ab88ab , archivePrefix =. 2001.10761 , primaryClass =

    Pith/arXiv arXiv 2001

  37. [37]

    , keywords =

    Characterizing the Fast Radio Burst Host Galaxy Population and its Connection to Transients in the Local and Extragalactic Universe. , keywords =. doi:10.3847/1538-3881/ac3aec , archivePrefix =. 2108.01282 , primaryClass =

    Pith/arXiv arXiv

  38. [38]

    , keywords =

    Fast radio burst source properties and curvature radiation model. , keywords =. doi:10.1093/mnras/stx665 , archivePrefix =. 1703.06139 , primaryClass =

    Pith/arXiv arXiv

  39. [39]

    , keywords =

    Fast radio bursts as synchrotron maser emission from decelerating relativistic blast waves. , keywords =. doi:10.1093/mnras/stz700 , archivePrefix =. 1902.01866 , primaryClass =

    Pith/arXiv arXiv 1902

  40. [40]

    , keywords =

    A bright millisecond-duration radio burst from a Galactic magnetar. , keywords =. doi:10.1038/s41586-020-2863-y , archivePrefix =. 2005.10324 , primaryClass =

    Pith/arXiv arXiv 2005

  41. [41]

    , keywords =

    FRBCAT: The Fast Radio Burst Catalogue. , keywords =. doi:10.1017/pasa.2016.35 , archivePrefix =. 1601.03547 , primaryClass =

    Pith/arXiv arXiv 2016

  42. [42]

    , keywords =

    Fast radio bursts at the dawn of the 2020s. , keywords =. doi:10.1007/s00159-022-00139-w , archivePrefix =. 2107.10113 , primaryClass =

    Pith/arXiv arXiv

  43. [43]

    , keywords =

    A fast radio burst associated with a Galactic magnetar. , keywords =. doi:10.1038/s41586-020-2872-x , archivePrefix =. 2005.10828 , primaryClass =

    Pith/arXiv arXiv 2005

  44. [44]

    arXiv e-prints , keywords =

    Unveiling the spectral morphological division of fast radio bursts with CHIME/FRB Catalog 2. arXiv e-prints , keywords =. doi:10.48550/arXiv.2601.16048 , archivePrefix =. 2601.16048 , primaryClass =

    work page doi:10.48550/arxiv.2601.16048

  45. [45]

    , keywords =

    A Practical Framework for Estimating the Repetition Likelihood of Fast Radio Bursts from Spectral Morphology. , keywords =. doi:10.3847/1538-4357/ae3c85 , archivePrefix =. 2510.16338 , primaryClass =

    arXiv

  46. [46]

    and Sridhar, Navin and Margalit, Ben and Beniamini, Paz and Sironi, Lorenzo , year=

    Metzger, Brian D. and Sridhar, Navin and Margalit, Ben and Beniamini, Paz and Sironi, Lorenzo , year=. A Toy Model for the Time–Frequency Structure of Fast Radio Bursts: Implications for the CHIME/FRB Burst Dichotomy , volume=. The Astrophysical Journal , publisher=. doi:10.3847/1538-4357/ac3b4a , number=

    work page doi:10.3847/1538-4357/ac3b4a

  47. [47]

    Latent Space Characterization of Autoencoder Variants

    Latent Space Characterization of Autoencoder Variants. arXiv e-prints , keywords =. doi:10.48550/arXiv.2412.04755 , archivePrefix =. 2412.04755 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2412.04755

  48. [48]

    , keywords =

    The Second CHIME/FRB Catalog of Fast Radio Bursts. , keywords =. doi:10.3847/1538-4365/ae3828 , archivePrefix =. 2601.09399 , primaryClass =

    arXiv

  49. [49]

    arXiv e-prints , keywords =

    Using Deep Learning for Robust Classification of Fast Radio Bursts. arXiv e-prints , keywords =. doi:10.48550/arXiv.2511.02634 , archivePrefix =. 2511.02634 , primaryClass =

    work page doi:10.48550/arxiv.2511.02634

  50. [50]

    , keywords =

    Morphology of 137 Fast Radio Bursts Down to Microsecond Timescales from the First CHIME/FRB Baseband Catalog. , keywords =. doi:10.3847/1538-4357/ad9b11 , archivePrefix =. 2408.13215 , primaryClass =

    Pith/arXiv arXiv

  51. [51]

    , keywords =

    Modeling the Morphology of Fast Radio Bursts and Radio Pulsars with fitburst. , keywords =. doi:10.3847/1538-4365/ad27d6 , archivePrefix =. 2311.05829 , primaryClass =

    Pith/arXiv arXiv

  52. [52]

    The Astrophysical Journal Supplement Series , volume=

    Mitigating radio frequency interference in CHIME/FRB real-time intensity data , author=. The Astrophysical Journal Supplement Series , volume=. 2023 , publisher=

    2023

  53. [53]

    Learning Deep Representation for Imbalanced Classification , year=

    Huang, Chen and Li, Yining and Loy, Chen Change and Tang, Xiaoou , booktitle=. Learning Deep Representation for Imbalanced Classification , year=

  54. [54]

    , keywords =

    Revealing hidden repeaters in the CHIME/FRB catalogue: semisupervised insights into the fast radio burst population. , keywords =. doi:10.1093/mnras/staf2178 , archivePrefix =. 2512.06316 , primaryClass =

    arXiv

  55. [55]

    Deep Clustering with Convolutional Autoencoders

    Guo, Xifeng and Liu, Xinwang and Zhu, En and Yin, Jianping. Deep Clustering with Convolutional Autoencoders. Neural Information Processing. 2017

    2017

  56. [56]

    ICONIP17-DCEC

    A better autoencoder for image: Convolutional autoencoder , author=. ICONIP17-DCEC. Available online: http://users. cecs. anu. edu. au/Tom. Gedeon/conf/ABCs2018/paper/ABCs2018\_paper\_58. pdf (accessed on 23 March 2017) , pages=

    2017

  57. [57]

    Machine learning classification of CHIME fast radio bursts – I

    Luo, Jia-Wei and Zhu-Ge, Jia-Ming and Zhang, Bing , year=. Machine learning classification of CHIME fast radio bursts – I. Supervised methods , volume=. Monthly Notices of the Royal Astronomical Society , publisher=. doi:10.1093/mnras/stac3206 , number=

    doi:10.1093/mnras/stac3206

  58. [58]

    Monthly Notices of the Royal Astronomical Society , volume =

    Zhu-Ge, Jia-Ming and Luo, Jia-Wei and Zhang, Bing , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2022 , month =. doi:10.1093/mnras/stac3599 , url =

    doi:10.1093/mnras/stac3599 2022

  59. [59]

    Monthly Notices of the Royal Astronomical Society , volume =

    Chen, Bo Han and Hashimoto, Tetsuya and Goto, Tomotsugu and Raquel, Bjorn Jasper R and Uno, Yuri and Kim, Seong Jin and Hsiao, Tiger Y-Y and Ho, Simon C-C , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2023 , month =. doi:10.1093/mnras/stad930 , url =

    work page doi:10.1093/mnras/stad930 2023

  60. [60]

    Monthly Notices of the Royal Astronomical Society , volume =

    Chen, Bo Han and Hashimoto, Tetsuya and Goto, Tomotsugu and Kim, Seong Jin and Santos, Daryl Joe D and On, Alvina Y L and Lu, Ting-Yi and Hsiao, Tiger Y-Y , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2021 , month =. doi:10.1093/mnras/stab2994 , url =

    doi:10.1093/mnras/stab2994 2021

  61. [61]

    Monthly Notices of the Royal Astronomical Society , volume =

    Raquel, Bjorn Jasper R and Hashimoto, Tetsuya and Goto, Tomotsugu and Chen, Bo Han and Uno, Yuri and Hsiao, Tiger Yu-Yang and Kim, Seong Jin and Ho, Simon C-C , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2023 , month =. doi:10.1093/mnras/stad1942 , url =

    work page doi:10.1093/mnras/stad1942 2023

  62. [62]

    Monthly Notices of the Royal Astronomical Society , volume =

    Sharma, Arjun and Rajpaul, Vinesh Maguire , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2024 , month =. doi:10.1093/mnras/stae1972 , url =

    work page doi:10.1093/mnras/stae1972 2024

  63. [63]

    , author=

    hdbscan: Hierarchical density based clustering. , author=. J. Open Source Softw. , volume=

  64. [64]

    and Varoquaux, G

    Pedregosa, F. and Varoquaux, G. and Gramfort, A. and Michel, V. and Thirion, B. and Grisel, O. and Blondel, M. and Prettenhofer, P. and Weiss, R. and Dubourg, V. and Vanderplas, J. and Passos, A. and Cournapeau, D. and Brucher, M. and Perrot, M. and Duchesnay, E. , journal=. Scikit-learn: Machine Learning in

  65. [65]

    and Siemion, A

    Gajjar, V. and Siemion, A. P. V. and Price, D. C. and Law, C. J. and Michilli, D. and Hessels, J. W. T. and Chatterjee, S. and Archibald, A. M. and Bower, G. C. and Brinkman, C. and Burke-Spolaor, S. and Cordes, J. M. and Croft, S. and Enriquez, J. Emilio and Foster, G. and Gizani, N. and Hellbourg, G. and Isaacson, H. and Kaspi, V. M. and Lazio, T. J. W....

    2018