Pith. sign in

REVIEW 3 major objections 5 minor 20 references

MARS: A Lightweight Morphology-Aware RFI Segmentation Network for Mask-Guided Mitigation in Radio Astronomy

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

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

Pith's one-line read A compact 270,769-parameter morphology-aware U-Net segments radio frequency interference with 0.978 F1 and 0.995 precision, preserves 97.6% of injected dispersed-signal fluence, and runs 6.2–7.0x faster than filtool in compute time.

desk verdict A careful, well-scoped RFI-mitigation paper whose headline numbers are still in-distribution with its synthetic training family—worth refereeing, but real-world transfer claims need artifacts and more data. read the letter →

arxiv 2608.05546 v1 pith:I7JDNKQC submitted 2026-08-06 astro-ph.IM

Zhaocheng Gong , Jack White , Jayanta Roy , Wesley Armour This is my paper · ORCID
classification astro-ph.IM
keywords radiofrequencyinterferenceRFImitigationpulsarsearchfastburstssegmentationnetworkU-NetGPUcomputingfilterbankprocessing
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 RFI mitigation for pulsar and fast radio burst searches can be done accurately and fast on a GPU by a compact, morphology-aware segmentation network rather than a large generic backbone or a CPU tool. The proposed MARS pipeline predicts full-resolution RFI masks, replaces masked samples in the original filterbank data, and rescales the output for downstream searches. The authors report near-perfect mask quality on synthetic patches (F1 0.978, precision 0.995), high preservation of injected dispersed signals (97.6% fluence in clean patches), PRESTO candidate significance close to the filtool baseline, and a 6.2–7.0x compute-only speedup. If these controlled results transfer to real observations, MARS would make near-real-time GPU-resident RFI mitigation practical for next-generation arrays while protecting the dispersed transients that searches are designed to find.

What carries the argument

The load-bearing object is the shape-aware bottleneck and decoder refinement: three parallel convolutional branches (local 3x3, horizontal 1x9, vertical 9x1) on the deepest feature map, plus horizontal and vertical residual refinement blocks (1x31 or 31x1 followed by 3x3) at each decoder scale, giving an explicit inductive bias for compact, time-extended, and frequency-extended interference without a large backbone. The second key mechanism is the astronomical-signal preservation loss, a quadratic penalty on high RFI probability over clean injected-pulse pixels, which is what raises retained fluence from 0.888 to 0.976 in clean patches. Around these sits a GPU-resident processing chain: per-channel median and median-absolute-deviation normalisation with tanh compression, 512x512 patch inference, mask reconstruction with channel-occupancy promotion, zero replacement in a mean/std-normalised stream, baseline correction, and 8-bit rescaling, all executed on GPU with TensorRT FP16 inference.

What would settle it

Take real filterbank observations containing RFI with steeply drifting or curved frequency-time tracks, label the RFI by careful manual inspection or an independent method, and measure MARS's patch-level F1 and retained fluence on those patches. If F1 falls well below the 0.978 synthetic number, or if a bright dispersed pulse overlapped by such RFI loses most of its fluence, the central generalization claim fails; with only two real GMRT observations, this is currently untested.

Watch

Extended reading notes

Core claim

The central claim is that the recurring local and anisotropic time-frequency morphology of RFI in filterbank data can be captured by a small deployment-oriented network, and that the resulting masks, applied through an explicit replacement pipeline, preserve the science content of pulsar and FRB searches. Concretely, a 270,769-parameter full-resolution U-Net with a shape-aware bottleneck achieves 0.978 F1 and 0.995 precision on SPECTRALib synthetic RFI, retains 97.6% of injected dispersed-signal fluence in clean patches and 96.4% in mixed-RFI patches, yields median PRESTO target-significance ratios of 0.90–0.99 relative to filtool on synthetic binary-pulsar filterbanks, and recovers the known pulsars in two GMRT observations. The drop to 0.888 and 0.837 retained fluence when the astronomical-signal preservation loss is removed identifies that loss term as the mechanism protecting compact, bright, low-dispersion-measure pulses.

Load-bearing premise

The load-bearing premise is that SPECTRALib-style synthetic RFI, which consists of additive, axis-aligned, constant-amplitude regions clipped to 8-bit, represents the RFI that real telescopes produce; the paper itself concedes this does not cover every drifting, curved, stochastic, or instrument-specific morphology present in real observations.

Editorial extensions

If this is right

  • RFI mitigation can move inside GPU pulsar and FRB search pipelines: the full mitigation computation takes 0.34–3.31 seconds for a roughly 100-second, 4096-channel filterbank, a 6.2–7.0x compute-only reduction versus the fastest filtool configuration tested.
  • Dispersed transients are preserved much better than with a generic segmentation baseline: 97.6% and 96.4% injected fluence retained for clean and mixed-RFI patches, versus 57.7% and 53.6% for the RFDL baseline.
  • Cleaned filterbanks remain search-compatible: period-matched PRESTO candidates recover median significance ratios of 0.90–0.99 relative to filtool, approaching equality at 1024 and 1310 microseconds sampling times.
  • The astronomical-signal preservation loss is a necessary component, not a nicety: removing it drops retained fluence to 0.888 in clean patches and 0.837 in mixed-RFI patches, with the largest loss on low-DM, high-S/N pulses.
  • Real-data checks pass in two GMRT observations, recovering the known pulsars with comparable or slightly higher significance than filtool.

Reading between the lines

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

  • Editorial: the same architecture-and-loss recipe should generalize to other telescopes if retrained on their RFI statistics; the design principle of small anisotropic morphology biases plus an explicit penalty on flagging dispersed power is not tied to GMRT or to SPECTRALib.
  • Editorial: a concrete testable extension is to train on synthetic RFI with drifting or curved frequency-time tracks and evaluate F1 and fluence retention; the paper's own limitation note predicts that performance would depend on how far real RFI departs from the axis-aligned family.
  • Editorial: the lower-tail PRESTO losses at 128–512 microseconds point to the replacement policy, not the mask, as the next bottleneck; replacing zero-fill with local interpolation or inpainting from adjacent channels could recover some lost significance.
  • Editorial: at the measured compute rates, streaming I/O rather than compute is likely to become the limiting factor for online mitigation, so the practical next test is an end-to-end online benchmark that includes disk and network transfer.
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. The paper presents MARS, a GPU-oriented RFI mitigation pipeline for pulsar and FRB search workflows. The neural component is a compact 270k-parameter U-Net with a morphology-aware bottleneck (local, horizontal, and vertical convolutional branches) designed to segment RFI in the frequency-time plane, and the full pipeline implements normalisation, patch inference, mask reconstruction, replacement, baseline removal, and rescaling on GPU with TensorRT-accelerated inference. Training uses synthetic patches with a weighted BCE + Dice + astronomical-preservation loss (L_astro), and evaluation is staged at three levels: a 15,000-patch SPECTRALib RFI-mask benchmark versus a retrained RFDL baseline; a 18,720-patch dispersed-pulse fluence-retention benchmark with and without mixed injected RFI; and a filterbank-level comparison against filtool via matched PRESTO searches on synthetic binary-pulsar files and two real GMRT observations. Reported headline results are RFI-mask F1 of 0.978, precision 0.995, retained fluence fractions of 0.976/0.964, PRESTO significance ratios of 0.90-0.99, and compute-only speedups of 6.2-7.0x over the fastest tested filtool configuration.

Significance. If the transfer to real RFI is established, the contribution is significant: the architecture is genuinely lightweight and TensorRT-deployable, the L_astro preservation loss is a useful and well-motivated design, and the ablation models are retrained from scratch rather than disabled at inference time. The evaluation protocol is carefully staged, the threshold tau=0.5 is fixed before all reported evaluations, the reported arithmetic is internally consistent (F1=0.978 with precision 0.995 and recall 0.963), and the paper is unusually candid in listing its own limitations. The main unresolved question is whether the SPECTRALib-derived quantitative results, which dominate the abstract and conclusions, transfer to real RFI; the current real-data evidence is too thin to support deployment-ready claims.

major comments (3)
  1. [Section 3, Section 4.1, Appendix E] The headline patch-level numbers are near-in-distribution measurements rather than evidence of real-RFI coverage. The training augmentation in Section 3 injects horizontal, vertical, compact, block-like, periodic, and composite RFI morphologies, while Appendix E says that SPECTRALib's RFI model is based on additive axis-aligned constant-amplitude regions clipped to 8-bit and that the training augmentation is 'SPECTRALib-style' implemented in the normalised input domain. Section 4.1 then evaluates on SPECTRALib-generated patches from the same morphology families. Consequently, F1=0.978, precision=0.995, and the fluence-retention values 0.976/0.964 measure how well the model fits its own training distribution, not how well it covers real drifting, curved, stochastic, or instrument-specific RFI. Because the paper's practical claim of near-real-time deployment rests on transfer beyond this family, this is load-bearing. Please add an out-of-distribution evaluation, for example synthetic RFI with curved, drifting, or non-axis-aligned structures, or real observations with independently obtained ground-truth masks, and report the resulting F1 and fluence gap; alternatively, explicitly re-scope the quantitative claims as simulator-benchmark results.
  2. [Section 4.3 and Section 5.3] The PRESTO significance comparison is conditional on the target candidate being detected in both cleaned outputs. The text states that the significance comparison includes only realizations in which the injected target candidate is detected in both the filtool- and GPU-NN-cleaned files. The reported median significance ratios (0.946, 0.913, 0.898, 0.982, and 0.988 at the five sampling times) therefore cannot reveal cases where the GPU pipeline suppresses the candidate below the search threshold while filtool detects it. Please report the per-method detection counts out of 200 for each sampling time (Figure 12 gives n but not the detection fraction) and provide an unconditional comparison that treats non-detections explicitly, for example through detection-rate ratios or sensitivity upper limits.
  3. [Section 5.4 and Table 7] The real-data validation is limited to two bright, known-DM, non-accelerated GMRT pulsar searches, with recovered PRESTO significances of 13.89 and 17.09. These tests show that a strong periodic signal survives mitigation, but they cannot bound real-RFI mask accuracy, cannot measure false-positive masking of faint dispersed pulses, and cannot support generalisation across telescopes, observing bands, or RFI environments. A concrete strengthening would be to inject faint dispersed pulses with controlled DM and S/N into these real observations before mitigation and measure recovered S/N and fluence as a function of proximity to real RFI, and to report the flagging fraction and mask occupancy of the MARS output on real data relative to filtool. Without such evidence, the real-data section remains an existence check rather than validation of the central real-world generalization claim.
minor comments (5)
  1. [Section 2.2] The phrase 'minimum valid scaled min' should read 'minimum valid scaled median absolute deviation' for consistency with the definition of d_c,s.
  2. [Sections 2-5] The name 'MARS' is used both for the neural mask predictor and for the complete GPU mitigation pipeline, which is elsewhere called 'GPU RFI mitigation pipeline'; please define the naming once and use it consistently, since the performance claims apply to different entities.
  3. [Section 4.2] The retained-fluence metric excludes injected-signal pixels that overlap ground truth RFI and limits overlap to at most 15% of the signal support; this is transparent in the text, but reporting the actually realised mean overlap fraction would help readers interpret the 0.964 figure.
  4. [Section 4.6] The runtime comparison is compute-only and compares different hardware (GH200 GPU versus EPYC CPU); the text is explicit about this, but the conclusion would benefit from a sentence restating that the 6.2-7.0x figure is not a hardware-independent algorithmic speedup.
  5. [General] No code, trained checkpoint, or model artifact is released, which prevents independent reproduction of the central transfer claims; please add a data/code availability statement or a reproducibility appendix with exact artifact URLs and versioned hashes.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: headline patch-level numbers are explicitly controlled simulator benchmarks, and the pipeline-level claims are anchored by independent filtool PRESTO comparisons and two real GMRT pulsar recoveries.

full rationale

The derivation chain is not circular. Training uses SPECTRALib-style synthetic augmentation (Section 3: 'Synthetic RFI structures with horizontal, vertical, compact, block-like, periodic, and composite morphologies are injected'), and the patch-level benchmark uses SPECTRALib (Section 4.1: 'This evaluation uses SPECTRALib to inject diverse RFI'), so the F1=0.978 and fluence-retention numbers are in-distribution with respect to the training morphology family. However, the paper explicitly labels these as controlled tests and does not present them as real-data F1; Appendix E concedes that the synthetic family 'does not represent every drifting, curved, stochastic, or instrument-specific morphology present in real observations.' The central transferable claims are anchored externally: the filterbank-level PRESTO comparison against filtool uses matched settings on independently generated SIGPROC binary-pulsar filterbanks, and two real GMRT observations recover known pulsars at comparable significance. The L_astro ablation demonstrates a design-intended effect (a loss term that penalizes RFI probability on injected pulse pixels improves fluence retention), which is a sanity check rather than an emergent prediction. No equation-level identity, fitted-parameter rename, or load-bearing uniqueness/self-citation chain forces any reported result. The main caveats are limited real-data validation and the in-distribution nature of the synthetic benchmark, which are validity and transfer concerns, not circularity.

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

The pipeline's numerical results depend on a set of hand-set thresholds and loss weights, listed as free parameters. The scientific relevance of the numbers depends on domain assumptions: SPECTRALib synthetic RFI approximates real RFI morphology, injected dispersed pulses cover the relevant FRB and single-pulse parameter space, and the replacement policy plus PRESTO candidate matching is a valid measure of search sensitivity. These assumptions are stated in the paper, and Section 6.3 acknowledges the simulation-to-real gap. No new physical entities are introduced.

free parameters (6)
  • tanh compression scale alpha_tanh = 6
    Hand-set in Section 2.2; controls dynamic range compression of normalized patches and therefore affects all mask predictions.
  • minimum scale threshold d_min = 0.5
    Hand-set in Section 2.2 Eq. (3); determines when the per-channel MAD scale is replaced by the standard deviation.
  • sigmoid threshold tau = 0.5
    Fixed in Section 2.4; all reported F1, precision, recall, and retained-fluence numbers use this threshold.
  • channel occupancy threshold rho = 0.5
    Used in Section 2.3 to promote high-occupancy channels to fully flagged; affects replacement and downstream PRESTO significance.
  • positive-class weight w+ = 3.0
    Chosen in Section 3 to balance RFI and background pixel imbalance; shifts the precision-recall tradeoff.
  • loss weights lambda_Dice and lambda_astro = 1.0 and 1.0; 0 in the no-L_astro ablation
    Section 3. The ablation shows lambda_astro changes clean-patch retained fluence from 0.888 to 0.976, so the central signal-preservation claim depends on this choice.
assumptions (5)
  • domain assumption Synthetic SPECTRALib RFI, generated as additive constant-amplitude axis-aligned regions clipped to 8-bit, is representative of real RFI morphologies.
    Training augmentation uses SPECTRALib-style morphology and patch-level tests use SPECTRALib directly; Appendix E acknowledges this does not represent every drifting, curved, stochastic, or instrument-specific morphology.
  • domain assumption Injected dispersed FRB and single-pulse signals with DM 10-3000 pc cm^-3, S/N 6-20, and widths 1.31-15.72 ms cover the signal space the network must protect.
    Section 4.2 defines the protection benchmark on this grid; real FRB diversity is wider.
  • domain assumption Mask-guided zero replacement with baseline removal and 8-bit rescaling is a valid cleaning policy comparable to filtool for downstream PRESTO searches.
    Section 2.3 and 4.5; the filterbank-level comparison assumes both cleaned outputs are search-compatible.
  • domain assumption PRESTO matched-candidate significance on SIGPROC synthetic binary-pulsar files is a sufficient proxy for search sensitivity.
    Section 4.3 uses only candidates detected in both outputs; weak signals not recovered by either method are excluded.
  • domain assumption Per-channel median and MAD normalization with tanh compression preserves the frequency-time RFI morphology the network needs to see.
    Section 2.2 Eq. (5); the network sees only normalized and compressed patches.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MARS: A Lightweight Morphology-Aware RFI Segmentation Network for Mask-Guided Mitigation in Radio Astronomy." pith.science (2026). https://pith.science/paper/I7JDNKQC

@misc{pith2026260805546,
  author       = {Pith},
  title        = {Pith review of: MARS: A Lightweight Morphology-Aware RFI Segmentation Network for Mask-Guided Mitigation in Radio Astronomy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I7JDNKQC}},
  note         = {Machine review of arXiv:2608.05546}
}
abstract

Next-generation radio telescopes generate filterbank data at rates that make storing all observations for later offline mitigation impractical. Mitigation must therefore operate in real or near-real time within the search pipeline while preserving dispersed astrophysical signals. CPU tools fit GPU-centred search pipelines poorly, while neural alternatives can be computationally heavy. We present MARS, a GPU-based RFI mitigation pipeline centred on a lightweight Morphology-Aware RFI Segmentation Network. The model is a reduced-width, full-resolution U-Net with a bottleneck containing local, horizontal, and vertical filters to capture compact and elongated RFI structures in the frequency-time plane. Normalisation, patch construction, mask reconstruction, replacement, baseline removal, and output rescaling are also implemented on GPU. Training includes an astronomical-signal preservation loss that discourages false flagging of dispersed pulses. In controlled patch-level tests, MARS achieves an RFI-mask F1 score of $0.978$ and a precision of $0.995$. It retains $97.6\%$ of the injected dispersed-signal fluence in clean patches and $96.4\%$ of the non-overlapping signal fluence in patches containing mixed injected RFI. Ablation experiments show that the astronomical-signal preservation loss particularly improves the protection of compact, low-DM, high-S/N pulses. At filterbank level, period-matched PRESTO candidates recovered after MARS mitigation have median significance ratios of $0.90$--$0.99$ relative to filtool. Both methods also recover the known pulsars in two real GMRT observations. On an NVIDIA GH200 GPU, MARS achieves a compute-only speedup of $6.2\times$--$7.0\times$ over the fastest tested multi-threaded filtool configurations on an AMD EPYC 9825 CPU.

Figures

Figures reproduced from arXiv: 2608.05546 by the authors.

Figure 1
Figure 1. Representative training samples after data augmentation. Each 512 × 512 frequency-time patch was generated from a non-empty training-set background using the same augmentation configuration employed during MARS training. Panel (a) contains a diffuse persistent bright band and vertical bursts; panel (b) contains rectangular blocks and persistent narrowband RFI; and panel (c) contains sparse bright horizontal bars and… view at source ↗
Figure 2
Figure 2. Training and validation loss of MARS over 50 epochs. Both losses decrease rapidly during the initial epochs and subsequently stabilise, indicating convergence within the adopted training budget. The dashed vertical line marks the checkpoint selected for the main experiments, corresponding to the highest validation F1 score (0.9962) at epoch 49. 10−6 . This validation-driven schedule avoided prescribing a fixed decay… view at source ↗
Figure 3
Figure 3. Overall patch-level RFI detection performance on the fixed-seed synthetic RFI benchmark. MARS achieves higher precision and F1 score than the RFDL baseline when both models are evaluated using the same test patches, threshold, and metrics. baseline, its timing is reported separately from the full GPU RFI mitigation pipeline timing. For filtool, the reported compute time includes the mitigation computation performed … view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Patch-level RFI-detection performance of MARS across the six synthetic RFI families, reported in terms of precision, recall, and F1 score [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Patch-level RFI-detection performance of RFDL across the same six synthetic RFI families, reported in terms of precision, recall, and F1 score. 5. RESULTS 5.1. Patch-level RFI detection We first evaluate patch-level RFI detection on the fixed-seed synthetic RFI benchma…
Figure 6
Figure 6. Figure 6: Patch-level RFI F1 score as a function of sampling time and RFI family. Left: MARS. Right: RFDL baseline. MARS remains more stable across narrowband and mixed-RFI families, while RFDL degrades strongly for persistent and repeating narrowband structures. decreases to 0.…
Figure 7
Figure 7. Figure 7: Astrophysical-signal protection of MARS as a function of DM and S/N. The left panel shows the retained fluence fraction for clean injected FRB/single-pulse signals. The right panel shows the reduction in retained fluence caused by the addition of mixed RFI, defined as …
Figure 8
Figure 8. Figure 8: Astrophysical-signal protection of the RFDL baseline, evaluated on the same fixed-seed FRB/single-pulse benchmark as MARS. The left panel shows the retained fluence fraction for clean injected signals. The right panel shows the reduction in retained fluence after addin…
Figure 9
Figure 9. Figure 9: Patch-level FRB fluence retention for MARS and RFDL, evaluated on clean FRBs and FRBs with mixed injected RFI [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: Effect of the astronomical-signal preservation loss. The production model with λastro = 1 retains substantially more injected pulse fluence than the no-Lastro ablation in both clean-FRB patches and FRB+mixed-RFI patches [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Clean-FRB retained fluence with and without the astronomical-signal preservation loss. The largest gain from λastro = 1 occurs for compact, bright dispersed signals at low DM and high S/N, where the no-Lastro model is most likely to confuse injected pulse power with R…
Figure 12
Figure 12. Figure 12: PRESTO significance of matched target candidates recovered from synthetic contaminated filterbanks processed with filtool and the GPU RFI-mitigation pipeline. Each point represents one fixed-seed filterbank realisation in which the injected target signal was detected …
Figure 13
Figure 13. Figure 13: Distribution of the difference in recovered target-candidate significance between the GPU RFI-mitigation pipeline and filtool at each sampling time. For candidates matched to the injected target period, we define ∆σ = σGPU − σfiltool. Positive values indicate higher r…
Figure 14
Figure 14. Figure 14: Compute-only runtime comparison among the GPU RFI mitigation pipeline, RFDL, and filtool. Left: absolute compute time for 4096-channel, ∼ 100 s filterbanks. Right: compute time normalised by the GPU RFI mitigation pipeline. The GPU RFI mitigation pipeline and filtool …
Figure 15
Figure 15. Figure 15: Summary of the decoder-refinement ablation study. The panels compare aggregate patch-level RFI F1, retained FRB fluence in patches containing mixed injected RFI, and RFI-only recall in FRB-plus-mixed-RFI patches. FRB and RFI metrics exclude pixels where the two inject…
Figure 16
Figure 16. Figure 16: Aggregate patch-level RFI F1 scores of the decoder-refinement ablation variants as functions of sampling time and RFI family. Each model is evaluated on the same family-stratified test set. The stratified results in [PITH_FULL_IMAGE:figures/full_fig_p025_16.png]
Figure 17
Figure 17. Figure 17: Astrophysical-signal protection of the decoder-refinement ablation variants. For each variant, the left panel shows retained fluence for clean injected FRB signals, while the right panel shows retained fluence when mixed SPECTRALib RFI is also injected. In the mixed-R…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 18 canonical work pages

  1. [1]

    2015, TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems, https://www.tensorflow.org/

    Abadi, M., Agarwal, A., Barham, P., et al. 2015, TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems, https://www.tensorflow.org/

    2015

  2. [2]

    Development of production-ready GPU data processing pipeline software for AstroAccelerate

    Carels, C., Ad´ amek, K., Novotn` y, J., & Armour, W. 2019, arXiv preprint arXiv:1912.07704

    work page Pith review arXiv 2019

  3. [3]

    Cordes, J., & McLaughlin, M. A. 2003, The Astrophysical Journal, 596, 1142 Du Toit, C. D., Grobler, T. L., & Ludick, D. J. 2024, Monthly Notices of the Royal Astronomical Society, 530, 613

    work page 2003

  4. [4]

    2001, Astronomy & Astrophysics, 378, 327

    Fridman, P., & Baan, W. 2001, Astronomy & Astrophysics, 378, 327

    work page 2001

  5. [5]

    2024, Monthly Notices of the Royal Astronomical Society, 529, 4719

    Gu, F., Hao, L., Liang, B., et al. 2024, Monthly Notices of the Royal Astronomical Society, 529, 4719

    work page 2024

  6. [6]

    2011, Astrophysics Source Code Library, ascl

    Lorimer, D. 2011, Astrophysics Source Code Library, ascl

    work page 2011

  7. [7]

    J., Carli, E., & Desvignes, G

    Men, Y., Barr, E., Clark, C. J., Carli, E., & Desvignes, G. 2023, Astronomy & Astrophysics, 679, A20

    work page 2023

  8. [8]

    Nieuwpoort, R. V. 2022, Monthly Notices of the Royal Astronomical Society, 516, 5367

    work page 2022

Show all 20 references
  1. [9]

    2022, Monthly Notices of the Royal Astronomical Society, 510, 1393 NVIDIA Corporation

    Morello, V., Rajwade, K., & Stappers, B. 2022, Monthly Notices of the Royal Astronomical Society, 510, 1393 NVIDIA Corporation. 2026, NVIDIA TensorRT Documentation,, https://docs.nvidia.com/deeplearning/ tensorrt/latest/index.html

    2022

  2. [10]

    2010, Astrophysics Source Code Library, ascl

    Offringa, A. 2010, Astrophysics Source Code Library, ascl

    2010

  3. [11]

    2010, Monthly Notices of the Royal Astronomical Society, 405, 155

    Offringa, A., De Bruyn, A., Biehl, M., et al. 2010, Monthly Notices of the Royal Astronomical Society, 405, 155

    2010

  4. [12]

    2012, Astronomy & astrophysics, 539, A95

    Offringa, A., Van De Gronde, J., & Roerdink, J. 2012, Astronomy & astrophysics, 539, A95

    2012

  5. [13]

    Nieuwpoort, R. V. 2024, BOOK OF, 24

    2024

  6. [14]

    2019, Advances in neural information processing systems, 32

    Paszke, A., Gross, S., Massa, F., et al. 2019, Advances in neural information processing systems, 32

    2019

  7. [15]

    2019, The Astronomy and Astrophysics Review, 27, 4

    Petroff, E., Hessels, J., & Lorimer, D. 2019, The Astronomy and Astrophysics Review, 27, 4

    2019

  8. [16]

    J., Wicenec, A., Bennamoun, M., & Dodson, R

    Pritchard, N. J., Wicenec, A., Bennamoun, M., & Dodson, R. 2025, Communications Physics

    2025

  9. [17]

    M., Eikenberry, S

    Ransom, S. M., Eikenberry, S. S., & Middleditch, J. 2002, The Astronomical Journal, 124, 1788

    2002

  10. [18]

    2015, in International Conference on Medical image computing and computer-assisted intervention, Springer, 234–241 sigpyproc developers

    Ronneberger, O., Fischer, P., & Brox, T. 2015, in International Conference on Medical image computing and computer-assisted intervention, Springer, 234–241 sigpyproc developers. 2026, sigpyproc: Pulsar and FRB Data Analysis Library, stable, GitHub software repository https://s...

    2015

  11. [19]

    2026, SPECTRALib: Synthetic Pulsar Emission, Contamination, and Transients Radio Astronomy Library,, GitHub software repository https://github.com/jack-white1/SPECTRALib

    White, J. 2026, SPECTRALib: Synthetic Pulsar Emission, Contamination, and Transients Radio Astronomy Library,, GitHub software repository https://github.com/jack-white1/SPECTRALib

    2026

  12. [20]

    2022, Monthly Notices of the Royal Astronomical Society, 513, 4787

    Yuan, M., Zhu, W., Zhang, H., et al. 2022, Monthly Notices of the Royal Astronomical Society, 513, 4787

    2022

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.