Pith. sign in

REVIEW 4 major objections 5 minor 65 references

GraViT: Transfer Learning with Vision Transformers and MLP-Mixer for Strong Gravitational Lens Discovery

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Fine-tuned ViT and MLP-Mixer models identify strong gravitational lenses at levels comparable to or better than convolutional baselines on a common test sample, with one MLP-Mixer recovering 132 of 138 real lens candidates.

desk verdict Useful transfer-learning benchmark for lens finding, but the headline claim that ViTs/MLP-Mixers match or beat CNNs is not statistically established. read the letter →

arxiv 2509.00226 v1 pith:KUUHHJPW submitted 2025-08-29 cs.CV astro-ph.GA

René Parlange , Juan C. Cuevas-Tello , Octavio Valenzuela , Omar de J. Cabrera-Rosas , Tomás Verdugo , Anupreeta More , Anton T. Jaelani This is my paper
classification cs.CVastro-ph.GA
keywords stronggravitationallensingvisiontransformerMLP-MixertransferlearningimageclassificationensemblepredictionHSCsurveyLSST
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 transfer learning from large-scale pretrained image models is enough to make Vision Transformers and MLP-Mixers competitive with—and in places better than—the convolutional networks that have dominated strong gravitational lens search. It reproduces the evaluation protocol of a previous systematic CNN comparison, fine-tuning ten architectures on two existing HSC-based training sets and testing them on the same twelve test sets. On most single-class test sets the fine-tuned transformers reach AUC-ROC and F1 scores comparable to or above the CNN references, and the best integrated-test configuration reaches 0.92 AUC-ROC against the 0.94 reference. The clearest independent evidence is the held-out set of 138 real lens candidates, of which a partially fine-tuned MLP-Mixer recovers 132. A key scope condition, which the paper itself acknowledges, is that high scores on the mock-rich test sets largely reflect each model learning the simulation pipeline it was trained on, so broad generalization claims rest on the small real-candidate sample.

What carries the argument

The load-bearing mechanism is the transfer-learning pipeline: HSC gri cutouts are remapped to RGB, resized to 224×224, normalized with ImageNet statistics, and fed to models pretrained on ImageNet; fine-tuning is applied at one of three depths (head only, half the layers, or all layers). The evaluation machinery is the common test sample of twelve sets built from real lenses, real candidates, and mocks. The paper's distinctive architecture is the MLP-Mixer, which alternates token-mixing and channel-mixing MLPs, has O(n) cost in the number of image patches, and achieves the best L2 recovery and the best integrated-set score among the newly evaluated models.

What would settle it

Take the A2 MLP-Mixer that recovered 132/138 L2 candidates and run it on a sample of confirmed strong lenses drawn from surveys or simulations not used in C21 or J24, matched to HSC-like imaging. If recall on that independent sample is far below 95.65 percent while mock-set recall stays near perfect, the transfer-learning gain is largely simulator recognition, not general lens detection.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that fine-tuning pretrained transformer-style models is a viable and scalable route to strong lens discovery. On the common test sample, fine-tuned ViT/MLP-Mixer models achieve AUC-ROC and F1 scores comparable to or exceeding the CNN values reported by the reference comparison for most single-class test sets and for the real-candidate subset L2. Only on the fully integrated test set l does the best new configuration, an MLP-Mixer trained with full fine-tuning on combined data, fall slightly below the 0.94 reference, reaching 0.92. The authors read this as evidence that global-context architectures boosted by large-scale pretraining can capture

Load-bearing premise

The load-bearing premise is that the common test sample measures real generalization: because the L3 and L4 mock sets come from the same simulation pipelines as the training sets, near-perfect scores on those sets may only show each model recognizing its own simulator, leaving the 138-image L2 real-candidate set as the main evidence for genuine lens-finding ability.

Editorial extensions

If this is right

  • Fine-tuned transformer and mixer models can replace CNNs in lens-search pipelines without a loss in detection quality on matched simulated data.
  • Partial fine-tuning often matches or exceeds full fine-tuning, so lens classifiers can be adapted with modest training budgets.
  • The MLP-Mixer's linear cost scaling makes it a natural fit for the large cutout volumes expected from LSST.
  • Models trained on one simulation pipeline fail on lenses produced by another, so survey-ready classifiers should be trained on the simulation matching the target survey, or on combined data.
  • Soft-voting ensembles of diverse architectures improve robustness on the hardest non-lens sets and can be formed without retraining the individual models.

Reading between the lines

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

  • I read the near-perfect scores on the L3 mock sets as measuring how well a model recognizes its own training simulator; the only clean evidence about real lenses in this paper is the 138-image L2 subset, so the comparative claim should be read with that scope.
  • A decisive next experiment would be to run the same fine-tuned models on confirmed lenses from an independent survey with comparable image quality and compare recall with the 95.65 percent L2 figure.
  • For LSST, a practical extension would be to increase MLP-Mixer patch size (its cost grows linearly, unlike self-attention) and train on a mix of current mocks plus real candidates, then use the soft-voting ensemble as a prescreener for human vetting.
  • The S experiment's finding that halving the C21 training sample barely changes performance suggests diminishing returns from adding more mocks of the same pipeline; diverse real data, not more simulation volume, is the likely bottleneck.
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 / 5 minor

Summary. The paper presents GraViT, a PyTorch pipeline for transfer learning applied to strong gravitational lens detection. The authors fine-tune ten pretrained vision-transformer variants (ViT, DeiT, CaiT, DeiT III, Swin, Twins-SVT, Twins-PCPVT, PiT, CvT) and MLP-Mixer on two training datasets (C21 and J24, and their combination, plus a reduced C21 set), and evaluate on the common test sample defined by More et al. (2024). Results are reported as AUC-ROC and F1 for 12 test-set configurations, with additional inference on the L2 subset of real candidates. The paper claims performance comparable to or exceeding convolutional baselines, discusses ensemble predictions, and provides inference-time complexity analysis. The code and fine-tuned models are made publicly available.

Significance. If the empirical claims were statistically robust, this would be a useful contribution to the growing literature on machine-learning-based lens finding, especially for LSST: it provides a reproducible transfer-learning pipeline, public models, and a comparison across many transformer architectures on a shared benchmark. The authors also reproduce and extend the systematic CNN comparison of More et al. (2024), which is valuable for the community. However, the central comparative claim is weakened by the absence of uncertainty quantification and by a best-of-many selection procedure that makes the headline numbers difficult to interpret. The paper also includes an internal inconsistency in the list of evaluated architectures. These issues are fixable and do not invalidate the overall approach, but they currently limit the strength of the conclusions that can be drawn.

major comments (4)
  1. [Section 4; Tables 3 and 4] The list of evaluated architectures is inconsistent. Section 4 states that ten architectures are evaluated (ViT, DeiT, CaiT, DeiT III, Swin, Twins-SVT, Twins-PCPVT, PiT, CvT, MLP-Mixer), yet Tables 3 and 4 report results for 'ResNet-18' as a trained model in several experiments (e.g., Table 3 d and i; Table 4 C3). It is therefore unclear whether ResNet-18 is an additional evaluated architecture, a reference model, or an error. This ambiguity affects the definition of the candidate set over which the 'best' results are selected and must be clarified before the results can be reproduced.
  2. [Section 5.1, Table 3] The 'best model' entries in Table 3 are maxima over a large grid: 10 architectures (or 11 including ResNet-18) × 12 training/fine-tuning settings, plus ensemble variants. Reporting the maximum of such a grid as the model's performance and comparing it against a single reference value introduces substantial winner's bias. Many entries are actually below the More et al. (2024) values (e.g., set a: 0.97 vs 0.98; set c: 0.96 vs 0.99; set l: 0.92 vs 0.94), so the statement in §5.1 that results are 'comparable to or exceeding the state of the art' is not strongly supported by the table. The authors should either report means and standard deviations over multiple seeds, apply a proper model-selection rule (e.g., validation-based), or explicitly frame Table 3 as an upper envelope rather than a fixed-model comparison.
  3. [Section 5.2, Table 4] The L2 inference results are statistically fragile. The difference between the top recall (A2 MLP-Mixer, 132/138) and the second-best (C3 ResNet-18, 131/138) is a single image. For n=138, the standard error of a proportion near 0.95 is about 2 percentage points, so this gap is fully consistent with sampling noise. Moreover, the values in Table 4 are maxima over ten architectures for each experiment, further increasing the selection bias. In addition, the decision threshold used to define a 'detection' is never stated; without this threshold the recall numbers are not reproducible. The authors should specify the threshold, provide confidence intervals (e.g., Wilson intervals), and report results from multiple training seeds for at least the key configurations.
  4. [Section 2.5 and Section 5.1; Table 1] The common test sample contains mock lens sets L3 and L4 that are generated by the same simulation pipelines used to build the C21 and J24 training data. Consequently, the near-perfect AUC-ROC values on test sets g-k (and the fact that 6000 of the 6181 lenses in set l come from L3/L4) largely measure a model's ability to recognize its own training simulator rather than generalization to new real lenses. The authors do acknowledge provenance dependence in the Conclusions and Appendix C, but the abstract and §5.1 frame the results more broadly. The only independent held-out evidence is the 138-image L2 subset and the L1/L2 combinations in sets a-f. The comparative claims should be explicitly scoped to these independent subsets, and the wording 'state of the art' should be tempered accordingly.
minor comments (5)
  1. [Abstract and Section 5.1] The abstract states that 'the MLP-Mixer ensemble with C3 reaches 0.92 on the fully integrated test set l', but Table 3 lists the MLP-Mixer (not the ensemble) for C3. The ensemble AUC for l in Table C1 is 0.88. Please clarify whether the 0.92 refers to the individual MLP-Mixer or to an ensemble.
  2. [Section 4.1] The metrics definitions (§4.1) do not specify the probability threshold used to compute F1 or to count 'detections' in Table 4. A threshold (e.g., 0.5 or the validation-optimal threshold) should be clearly stated, as different thresholds can change the recall values substantially.
  3. [Figure 10] The text says that marker sizes in Figure 10 are scaled by 'mean AUC-ROC', but the figure description does not specify over which test sets or experiments the mean is computed. Please state the averaging domain (e.g., all test sets a-l for the C3 experiment).
  4. [Section 4.2] The sentence 'any model not included (reference ResNet-18) simply does not contribute to the normalization factor N' is confusing. If ResNet-18 is a reference and not part of the ensemble, this is fine, but it again highlights the need to clarify the full set of trained models.
  5. [Table C1/C2] The ensemble tables C1 and C2 report a single value per cell with no measure of variability. Even if the authors do not run multiple seeds, a statement to that effect and a caution that these are single-run results would aid interpretation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: GraViT is an empirical benchmark study; its central comparisons rest on an external common test sample and prior published baselines, not on definitions or self-citations that force the results.

full rationale

The paper does not derive any quantity from another by definition. It fine-tunes ten pre-trained architectures on the public C21 and J24 training sets (Sections 2.3-2.4) and evaluates them on the common test sample from More et al. (2024) (Section 2.5). The reference AUC-ROC/F1 values in Tables 3, C1 and C2 are copied from that prior published benchmark, not computed from the present model assumptions, so the comparison is an external benchmark rather than a self-referential reduction. Although More and Jaelani are co-authors and some test mocks (L3/L4) originate from the same simulation pipelines as the training sets, the paper explicitly acknowledges this provenance dependence ('an outcome that reflects the experimental design of the common test sample') and the L2 real-candidate inference (Table 4) supplies a held-out independent sample. The best-of-configuration reporting in Tables 3 and 4 is a statistical limitation (unquantified selection over many models, no uncertainty estimates) but it is not a circular step: the reported scores are measured outputs, not fitted parameters renamed as predictions, and no uniqueness theorem or ansatz is imported from the authors' prior work to force the conclusion. The central claim is therefore self-contained as an empirical comparison, even if its statistical strength can be questioned.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

No physical laws are derived; the ledger records the empirical modeling assumptions on which the benchmark conclusions depend.

free parameters (1)
  • Global training hyperparameter set = LR 1e-4, weight decay 1e-2, stochastic depth 0.1, early stopping patience 20, augmentation probability 0.5
    Chosen once by hand and applied uniformly across all models and datasets; not tuned to the test sets, but comparative conclusions about fine-tuning depth are conditional on this schedule.
assumptions (4)
  • domain assumption ImageNet-pretrained weights transfer to three-band HSC gri cutouts after band remapping (g,r,i to B,G,R) and ImageNet normalization.
    Invoked in Section 3.2 and Section 4; if false, fine-tuned transformers would lose their advantage over CNNs trained from scratch.
  • domain assumption Simulated lenses from GLEE (C21) and SIMCT (J24) capture enough real HSC image morphology to train a generalizable classifier.
    Sections 2.3-2.4; all training uses these mocks, so real-lens performance depends on this realism assumption.
  • domain assumption L1 and L2 labels in the common test sample are correct enough to serve as ground truth.
    Section 2.5; L2 items are network-selected candidates, not spectroscopically confirmed lenses, so recall numbers could be inflated by false positives in the labels.
  • domain assumption Mock test sets L3 and L4 drawn from the same simulators as training are treated as independent generalization tests.
    Table 1 and Appendix C; this inherited benchmark design choice inflates scores for C21-trained models on L3 and J24-trained models on L4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GraViT: Transfer Learning with Vision Transformers and MLP-Mixer for Strong Gravitational Lens Discovery." pith.science (2026). https://pith.science/paper/KUUHHJPW

@misc{pith2026250900226,
  author       = {Pith},
  title        = {Pith review of: GraViT: Transfer Learning with Vision Transformers and MLP-Mixer for Strong Gravitational Lens Discovery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KUUHHJPW}},
  note         = {Machine review of arXiv:2509.00226}
}
read the original abstract

Gravitational lensing offers a powerful probe into the properties of dark matter and is crucial to infer cosmological parameters. The Legacy Survey of Space and Time (LSST) is predicted to find O(10^5) gravitational lenses over the next decade, demanding automated classifiers. In this work, we introduce GraViT, a PyTorch pipeline for gravitational lens detection that leverages extensive pretraining of state-of-the-art Vision Transformer (ViT) models and MLP-Mixer. We assess the impact of transfer learning on classification performance by examining data quality (source and sample size), model architecture (selection and fine-tuning), training strategies (augmentation, normalization, and optimization), and ensemble predictions. This study reproduces the experiments in a previous systematic comparison of neural networks and provides insights into the detectability of strong gravitational lenses on that common test sample. We fine-tune ten architectures using datasets from HOLISMOKES VI and SuGOHI X, and benchmark them against convolutional baselines, discussing complexity and inference-time analysis.

Figures

Figures reproduced from arXiv: 2509.00226 by the authors.

Figure 1
Figure 1. C21: mock lenses from Cañameras et al. (2021) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. C21: non-lenses from Cañameras et al. (2021). conducted with the HSC, and its 1.77 𝑑𝑒𝑔2 field-of-view optical camera (Hyper Suprime-Cam Subaru Strategic Program 2024). 2.2 GAMA09H The Galaxy And Mass Assembly (GAMA)5 project is designed to exploit the latest generation of ground-based and space-borne survey facilities to study cosmology, galaxy formation, and evolution. HSC covers six discrete fields, named after ov… view at source ↗
Figure 3
Figure 3. J24: mock lenses from Jaelani et al. (2024) [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: J24: non-lenses from Jaelani et al. (2024) [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: Common test sample (More et al. 2024). Panels: left (lenses) vs. right (non-lenses) for each test set labeled a to l. MNRAS 000, 1–15 (2025) [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Gravitational lens detection with the Vision Transformer (ViT) architecture (Dosovitskiy et al. 2021). Twins-SVT Twins-SVT (Chu et al. 2021) introduces spatially separable self￾attention to reduce computational cost while preserving global fea￾ture interactions. It bal…
Figure 7
Figure 7. Figure 7: Lens classification diagram based on the MLP-Mixer architecture (Tolstikhin et al. 2021). conduct a distinct evaluation based on the training data and the degree of fine-tuning. The symbols represent the experiments as follows: • A corresponds to data from C21 (Cañamer…
Figure 8
Figure 8. Figure 8: Example of augmented lens from C21 (Cañameras et al. 2021). Data augmentations applied to the training data loader are listed on [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Ensemble results for experiments A- S (C21), B (J24) and C (C21+J24). The degree of fine-tuning is depicted by transparency: a lighter hue indicates that only the classifier head is updated, mid-tone color represents retraining half of the architecture, and full opacit…
Figure 10
Figure 10. Figure 10: Parameter count vs. floating point operations (FLOPs). Markers are scaled by the mean AUC–ROC, training every layer with sets C21+J24 (C3). The MLP-Mixer has a good balance between classification performance, in this case AUC–ROC (size of the marker), and its computat…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

65 extracted references · 25 canonical work pages

  1. [1]

    Aihara H., et al., 2019, @doi [Publications of the Astronomical Society of Japan] 10.1093/PASJ/PSZ103 , 71, 114

    doi:10.1093/pasj/psz103 2019

  2. [2]

    Astropy Collaboration et al., 2013, @doi [ ] 10.1051/0004-6361/201322068 , http://adsabs.harvard.edu/abs/2013A

    doi:10.1051/0004-6361/201322068 2013

  3. [3]

    Astropy Collaboration et al., 2018, @doi [ ] 10.3847/1538-3881/aabc4f , https://ui.adsabs.harvard.edu/abs/2018AJ....156..123A 156, 123

    doi:10.3847/1538-3881/aabc4f 2018

  4. [4]

    Astropy Collaboration et al., 2022, @doi [ ] 10.3847/1538-4357/ac7c74 , https://ui.adsabs.harvard.edu/abs/2022ApJ...935..167A 935, 167

    doi:10.3847/1538-4357/ac7c74 2022

  5. [5]

    L., Kiros J

    Ba J. L., Kiros J. R., Hinton G. E., 2016, arXiv preprint arXiv:1607.06450

    arXiv 2016

  6. [6]

    J., Treu T., et al., 2020, @doi [ ] 10.1051/0004-6361/202038861 , 643, A165

    Birrer S., Shajib A. J., Treu T., et al., 2020, @doi [ ] 10.1051/0004-6361/202038861 , 643, A165

    doi:10.1051/0004-6361/202038861 2020

  7. [7]

    Birrer S., et al., 2024, @doi [Space Science Reviews] 10.1007/s11214-024-01079-w , 220, 48

    doi:10.1007/s11214-024-01079-w 2024

  8. [8]

    Ca \ n ameras R., et al., 2021, @doi [Astronomy & Astrophysics] 10.1051/0004-6361/202141758 , 653, L6

    doi:10.1051/0004-6361/202141758 2021

Show all 65 references
  1. [9]

    Cao J., Xu T., Deng Y., Deng L., Yang M., Liu Z., Zhou W., 2024, @doi [Astronomy & Astrophysics] 10.1051/0004-6361/202348544 , 683, A42

    2024 doi

  2. [10]

    Carion N., Massa F., Synnaeve G., Usunier N., Kirillov A., Zagoruyko S., 2020, @doi [Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)] 10.1007/978-3-030-58452-8_13 , 12346 LNCS, 213

    2020 doi

  3. [11]

    Chu X., Tian Z., Wang Y., Zhang B., Ren H., Wei X., Xia H., Shen C., 2021, Advances in Neural Information Processing Systems, 12, 9355

    2021

  4. [12]

    Comparat J., et al., 2023, @doi [Astronomy & Astrophysics] 10.1051/0004-6361/202245726 , 673, A122

    2023 doi

  5. [13]

    J., Li K., Fei-Fei L., 2009, @doi [2009 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2009] 10.1109/CVPR.2009.5206848 , pp 248--255

    Deng J., Dong W., Socher R., Li L. J., Li K., Fei-Fei L., 2009, @doi [2009 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2009] 10.1109/CVPR.2009.5206848 , pp 248--255

    2009

  6. [14]

    ( @eprint arXiv 2010.11929 )

    Dosovitskiy A., et al., 2021, in International Conference on Learning Representations (ICLR). ( @eprint arXiv 2010.11929 )

    2021 arXiv

  7. [15]

    Erben T., et al., 2013, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/MNRAS/STT928 , 433, 2545

    2013 doi

  8. [16]

    Gonzalez J., et al., 2025, Discovering Strong Gravitational Lenses in the Dark Energy Survey with Interactive Machine Learning and Crowd-sourced Inspection with Space Warps ( @eprint arXiv 2501.15679 ), https://arxiv.org/abs/2501.15679

    2025 arXiv

  9. [17]

    Grespan M., Thuruthipilly H., Pollo A., Lochner M., Biesiada M., Etsebeth V., 2024, Astronomy & Astrophysics, p. 28

    2024

  10. [18]

    He K., Zhang X., Ren S., Sun J., 2015, @doi [Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition] 10.1109/CVPR.2016.90 , 2016-December, 770

    2015 doi

  11. [20]

    Hildebrandt H., et al., 2012, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/J.1365-2966.2012.20468.X/2/MNRAS0421-2355-F10.JPEG , 421, 2355

    2012

  12. [21]

    J., Verma A., More A., Ca \ n ameras R., Jaelani A

    Holloway P., Marshall P. J., Verma A., More A., Ca \ n ameras R., Jaelani A. T., Ishida Y., Wong K. C., 2023, MNRAS, 000, 1

    2023

  13. [22]

    M., Paech K., Bonnett C., Seitz S., Weller J., 2015, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stv1551 , 452, 4183

    Hoyle B., Rau M. M., Paech K., Bonnett C., Seitz S., Weller J., 2015, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stv1551 , 452, 4183

    2015 doi

  14. [23]

    Huang G., Sun Y., Liu Z., Sedra D., Weinberger K. Q., 2016, @doi [Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)] 10.1007/978-3-319-46493-0 \_ 39 , 9908 LNCS, 646

    2016 doi

  15. [24]

    Huang X., Storfer C., Ravi V., et al., 2020, @doi [ ] 10.3847/1538-4357/ab7ffb , 894, 78

    2020 doi

  16. [25]

    Huang X., Storfer C., Gu A., et al., 2021, @doi [ ] 10.3847/1538-4357/abd62b , 909, 27

    2021 doi

  17. [26]

    W., Chen G

    Huang K. W., Chen G. C. F., Chang P. W., Lin S. C., Hsu C. J., Thengane V., Lin J. Y. Y., 2023, @doi [Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)] 10.1007/978-3-031-25056-9 \_ 10 , 13801 LNCS, 143

    2023 doi

  18. [27]

    Hyper Suprime-Cam Subaru Strategic Program 2024, Public Data Release, https://hsc-release.mtk.nao.ac.jp/doc/index.php/data/

    2024

  19. [28]

    C., Jaelani A

    Ishida Y., Wong K. C., Jaelani A. T., More A., 2024, @doi [Publications of the Astronomical Society of Japan] 10.1093/pasj/psae102 , 77, 105

    2024 doi

  20. [29]

    Jacobs C., Glazebrook K., Collett T., More A., McCarthy C., 2017, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/MNRAS/STX1492 , 471, 167

    2017 doi

  21. [30]

    Jacobs C., Collett T., Glazebrook K., et al., 2019a, @doi [ ] 10.3847/1538-4365/ab26b6 , 243

    doi

  22. [31]

    Jacobs C., Collett T., Glazebrook K., et al., 2019b, @doi [ ] 10.1093/mnras/stz272 , 484, 5330

    doi

  23. [32]

    T., More A., Wong K

    Jaelani A. T., More A., Wong K. C., Inoue K. T., Chao D. C. Y., Premadi P. W., Ca \ n ameras R., 2024, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/MNRAS/STAE2442 , 535, 1625

    2024 doi

  24. [33]

    Jia P., Sun R., Li N., Song Y., Ning R., Wei H., Luo R., 2022, @doi [The Astronomical Journal] 10.3847/1538-3881/aca1c2 , 165, 26

    2022 doi

  25. [34]

    L., Rodney S., Treu T., et al., 2023, @doi [ ] 10.1126/science.abh1322 , 380

    Kelly P. L., Rodney S., Treu T., et al., 2023, @doi [ ] 10.1126/science.abh1322 , 380

    2023 doi

  26. [35]

    P., Ba J

    Kingma D. P., Ba J. L., 2014, 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings

    2014

  27. [37]

    Loshchilov I., Hutter F., 2017, 7th International Conference on Learning Representations, ICLR 2019

    2017

  28. [38]

    R., Fekete G., Hogg D

    Lupton R., Blanton M. R., Fekete G., Hogg D. W., O’Mullane W., Szalay A., Wherry N., 2004, @doi [Publications of the Astronomical Society of the Pacific] 10.1086/382245/XML , 116, 133

    2004 doi

  29. [39]

    B., Meneghetti M., Avestruz C., et al., 2019, @doi [ ] 10.1051/0004-6361/201832797 , 625, A119

    Metcalf R. B., Meneghetti M., Avestruz C., et al., 2019, @doi [ ] 10.1051/0004-6361/201832797 , 625, A119

    2019 doi

  30. [40]

    More A., et al., 2016, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/MNRAS/STV1965 , 455, 1191

    2016 doi

  31. [41]

    More A., et al., 2024, MNRAS, 000, 1

    2024

  32. [42]

    pp 6453--6462 ( @eprint arXiv 1802.05751 )

    Parmar N., Vaswani A., Uszkoreit J., Kaiser L., Shazeer N., Ku A., Tran D., 2018, in 35th International Conference on Machine Learning, ICML 2018. pp 6453--6462 ( @eprint arXiv 1802.05751 )

    2018 arXiv

  33. [43]

    Pascale M., et al., 2025, @doi [The Astrophysical Journal] 10.3847/1538-4357/AD9928 , 979, 13

    2025 doi

  34. [44]

    E., Tortora C., Vernardos G., et al., 2019, @doi [ ] 10.1093/mnras/stz189 , 484, 3879

    Petrillo C. E., Tortora C., Vernardos G., et al., 2019, @doi [ ] 10.1093/mnras/stz189 , 484, 3879

    2019 doi

  35. [45]

    Ridnik T., Ben-Baruch E., Noy A., Zelnik-Manor L., 2021, arXiv e-prints

    2021

  36. [46]

    Rojas K., Savary E., Cl\'ement B., et al., 2022, @doi [ ] 10.1051/0004-6361/202142119 , 668, A73

    2022 doi

  37. [47]

    E., Hinton G

    Rumelhart D. E., Hinton G. E., Williams R. J., 1986, nature, 323, 533

    1986

  38. [48]

    J., Vernardos G., Collett T

    Shajib A. J., Vernardos G., Collett T. E., et al., 2024,

    2024

  39. [49]

    H., Taubenberger S., Inoue K

    Shu Y., Ca \ n ameras R., Schuldt S., Suyu S. H., Taubenberger S., Inoue K. T., Jaelani A. T., 2022, @doi [Astronomy & Astrophysics] 10.1051/0004-6361/202243203 , 662, A4

    2022 doi

  40. [50]

    Sovrasov V., 2018-2024, ptflops: a flops counting tool for neural networks in pytorch framework, https://github.com/sovrasov/flops-counter.pytorch

    2018

  41. [51]

    Srivastava N., Hinton G., Krizhevsky A., Sutskever I., Salakhutdinov R., 2014, The journal of machine learning research, 15, 1929

    2014

  42. [52]

    Steiner A., Kolesnikov A., Zhai X., Wightman R., Uszkoreit J., Beyer L., 2021, Trans. Mach. Learn. Res., 2022

    2021

  43. [53]

    H., Halkola A., 2010, @doi [Astronomy & Astrophysics] 10.1051/0004-6361/201015481 , 524, A94

    Suyu S. H., Halkola A., 2010, @doi [Astronomy & Astrophysics] 10.1051/0004-6361/201015481 , 524, A94

    2010 doi

  44. [54]

    H., et al., 2012, @doi [The Astrophysical Journal] 10.1088/0004-637X/750/1/10 , 750, 10

    Suyu S. H., et al., 2012, @doi [The Astrophysical Journal] 10.1088/0004-637X/750/1/10 , 750, 10

    2012 doi

  45. [55]

    V., 2019, 36th International Conference on Machine Learning, ICML 2019, 2019-June, 10691

    Tan M., Le Q. V., 2019, 36th International Conference on Machine Learning, ICML 2019, 2019-June, 10691

    2019

  46. [56]

    Thuruthipilly H., Zadrozny A., Pollo A., Biesiada M., 2022, @doi [Astronomy & Astrophysics] 10.1051/0004-6361/202142463 , 664

    2022 doi

  47. [57]

    Thuruthipilly H., Grespan M., Zadro\.zny A., 2024, @doi [AIP Conference Proceedings] 10.1063/5.0203317

    2024 doi

  48. [58]

    Tolstikhin I., et al., 2021, Advances in Neural Information Processing Systems, 29, 24261

    2021

  49. [59]

    Touvron H., Cord M., Douze M., Massa F., Sablayrolles A., J \' e gou H., 2020, Proceedings of Machine Learning Research, 139, 10347

    2020

  50. [60]

    pp 32--42, https://api.semanticscholar.org/CorpusID:232428161

    Touvron H., Cord M., Sablayrolles A., Synnaeve G., J'egou H., 2021, in 2021 IEEE/CVF International Conference on Computer Vision (ICCV). pp 32--42, https://api.semanticscholar.org/CorpusID:232428161

    2021

  51. [61]

    Touvron H., Cord M., J \' e gou H., 2022, @doi [Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)] 10.1007/978-3-031-20053-3 \_ 30 , 13684 LNCS, 516

    2022 doi

  52. [62]

    N., Kaiser L., Polosukhin I., 2017, CoRR, abs/1706.03762

    Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A. N., Kaiser L., Polosukhin I., 2017, CoRR, abs/1706.03762

    2017 arXiv

  53. [63]

    Z., Khabsa M., Fang H., Ma H., 2020, in arxiv preprint

    Wang S., Li B. Z., Khabsa M., Fang H., Ma H., 2020, in arxiv preprint. https://arxiv.org/abs/2006.04768v3

    2020 arXiv

  54. [64]

    Wightman R., 2019, PyTorch Image Models, https://github.com/rwightman/pytorch-image-models, @doi 10.5281/zenodo.4414861

    2019 doi

  55. [65]

    Wu H., Xiao B., Codella N., Liu M., Dai X., Yuan L., Zhang L., 2021, @doi [Proceedings of the IEEE International Conference on Computer Vision] 10.1109/ICCV48922.2021.00009 , pp 22--31

    2021

  56. [66]

    Yao J., Lin Y., Liao S.-M., Huang H.-J., Kuo W.-T., Hsuan O., Ou M., 2021, arXiv preprint arXiv:2110.01024

    2021 arXiv

  57. [67]

    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.stat...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.