arxiv: 2603.10146 · v3 · submitted 2026-03-10 · ⚛️ physics.ins-det · hep-ph

Recognition: no theorem link

Polarized Target Nuclear Magnetic Resonance Measurements with Deep Neural Networks

Devin Seay , Ishara P. Fernando , Dustin Keller

Authors on Pith no claims yet

Pith reviewed 2026-05-15 12:46 UTC · model grok-4.3

classification ⚛️ physics.ins-det hep-ph

keywords neural networkscontinuous-wave NMRpolarization metrologysignal denoisingpolarized targetsmachine learningnuclear magnetic resonancehigh-energy physics

0 comments

The pith

Deep neural networks extract and denoise continuous-wave NMR signals to reduce polarization fitting uncertainties.

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

The paper establishes that neural network models can be applied to continuous-wave NMR data for the first time to perform signal extraction and denoising in polarized target measurements. Conventional Q-meter methods suffer from noise, baseline drift, and fitting errors that degrade precision in real-time and offline analysis. The networks deliver lower uncertainties across simulated and experimental conditions. This directly supports more reliable polarization monitoring during nuclear and high-energy physics runs. The result raises the effective figure of merit for scattering experiments that depend on dynamically polarized targets.

Core claim

We report the first successful application of neural network architectures to continuous-wave NMR polarization metrology. By leveraging advanced machine learning techniques for signal extraction and denoising, we achieve a substantial reduction of fitting uncertainties under a variety of realistic simulated and experimental conditions. These improvements translate directly into more robust real-time (online) polarization monitoring and higher precision in subsequent offline analysis.

What carries the argument

Neural network architectures trained for signal extraction and denoising on CW-NMR spectra.

If this is right

More robust real-time polarization monitoring during scattering runs
Higher precision in offline analysis of target polarization data
Improved overall figure of merit for experiments using dynamically polarized targets
A new analysis toolset for NMR-based polarimetry across nuclear and high-energy physics

Where Pith is reading between the lines

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

The same networks might be retrained on pulsed-NMR data to broaden the method beyond constant-current CW operation.
Hybrid pipelines that blend conventional Q-meter output with neural-network corrections could be tested for incremental adoption.
Collecting larger libraries of real beam-on experimental spectra would allow direct validation of simulation-to-real transfer without new hardware.

Load-bearing premise

Neural network models trained or validated primarily on simulated data will generalize to real experimental conditions without introducing new systematic biases or overfitting to simulation-specific features.

What would settle it

A side-by-side test on a large collection of real experimental CW-NMR datasets in which the neural-network polarization values show no reduction in uncertainty or exhibit larger systematic offsets than conventional fitting would falsify the claim.

Figures

Figures reproduced from arXiv: 2603.10146 by Devin Seay, Dustin Keller, Ishara P. Fernando.

**Figure 1.** Figure 1: Schematic of the Q-meter circuitry[27]. voltage across the Q-meter can be expressed as Re{ui} = U R0 Re{Z}+Y[Re2{Z}+Im2{Z}] [(1+Y Re{Z}) 2 +Y 2 Im2{Z}] (2) where Y = 1 Ri + 1 R0 and is the coupling admittance of the resonator, Ri is the total impedance of the amplifier, which can be expressed as Ri = R1i +R2i where R1i is assumed to be purely resistive, R0 is the current limiting resistance. A circuitr… view at source ↗

**Figure 2.** Figure 2: Top: Example of a simulated Q-meter baseline [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Energy level structure of deuterium in a magnetic [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Simulated deuteron absorption lineshape showing [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Simulated proton NMR signal demonstrating a Voigt [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

**Figure 6.** Figure 6: Examples of applied variations in the simulated [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗

**Figure 8.** Figure 8: Abstract Architecture Diagram of model used [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗

**Figure 10.** Figure 10: Simulation of TE signal with “high level” of noise [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗

**Figure 11.** Figure 11: Loss/Validation Loss vs. Epoch for area model [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗

**Figure 12.** Figure 12: The CNN-based model developed for the low-polarization regime (0–2%) significantly outperforms the corresponding MLP-based architecture. When evaluated on noiseless test data, the CNN yields a mean residual of 3 × 10−5% with a residual standard deviation of 2.3 × 10−4%, demonstrating excellent intrinsic accuracy and precision for such a small scaled signal. Under realistic noise conditions with signal-to-… view at source ↗

**Figure 13.** Figure 13: Distribution of residuals for polarization of [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗

**Figure 12.** Figure 12: Distribution of polarization residuals for the range [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗

**Figure 15.** Figure 15: Distribution of residuals from the area model for a [PITH_FULL_IMAGE:figures/full_fig_p020_15.png] view at source ↗

**Figure 14.** Figure 14: Distribution of residuals in area model for the [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗

**Figure 17.** Figure 17: Example of DAE being used on a real experimental [PITH_FULL_IMAGE:figures/full_fig_p021_17.png] view at source ↗

**Figure 18.** Figure 18: Another example of DAE being used on a real [PITH_FULL_IMAGE:figures/full_fig_p022_18.png] view at source ↗

**Figure 19.** Figure 19: Application of ss-RF at the resonance position [PITH_FULL_IMAGE:figures/full_fig_p022_19.png] view at source ↗

read the original abstract

Continuous-wave Nuclear Magnetic Resonance (CW-NMR) operated in constant-current mode has served as a foundational technique for polarization measurement in solid-state dynamically polarized targets within nuclear and high-energy physics experiments for several decades, and it remains an essential tool. Conventional Q-meter-based phase-sensitive detection is critical for precise real-time determination of target polarization during scattering runs. However, the accuracy and reliability of these measurements are frequently compromised by elevated noise levels, baseline drift, and systematic uncertainties arising from signal isolation and fitting, ultimately degrading the overall experimental figure of merit. In this work, we report the first successful application of neural network architectures to continuous-wave NMR polarization metrology. By leveraging advanced machine learning techniques for signal extraction and denoising, we achieve a substantial reduction of fitting uncertainties under a variety of realistic simulated and experimental conditions. These improvements translate directly into more robust real-time (online) polarization monitoring and higher precision in subsequent offline analysis. By reducing analysis-induced uncertainty, the resulting methodology can improve the effective figure of merit for scattering experiments employing dynamically polarized targets and provides a new toolset for NMR-based polarimetry in high-energy and nuclear physics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper tries neural nets on CW-NMR for polarized target polarization but gives no numbers to show the claimed uncertainty drop actually happens on real data.

read the letter

The core point is that this is the first reported use of neural networks for continuous-wave NMR signal extraction and denoising in solid polarized targets. The authors frame it as a way to cut fitting uncertainties from noise and baseline drift that have long limited Q-meter performance in nuclear and high-energy scattering runs. That is genuinely new within the subfield, and the motivation is practical: better real-time and offline polarization values would raise the effective figure of merit for those experiments. They say the approach works on both simulated and experimental conditions, which at least shows they tested beyond pure theory. The writing is straightforward about the decades-old problem with conventional phase-sensitive detection. That part is useful for anyone who has dealt with these signals. The main weakness is the absence of concrete evidence. The abstract asserts a substantial reduction in uncertainties but supplies no error bars, no direct comparison against standard fitting routines on the same datasets, and no architecture or training details. Without those, it is impossible to tell whether the networks deliver real gains or simply reproduce simulation artifacts. The generalization step from simulated training data to actual experimental runs is especially thin; real coil responses, drift patterns, and noise correlations often differ from what simulations capture, and nothing in the provided text shows they checked for domain shift on held-out real runs. This paper is aimed at the small group of people who build and run dynamically polarized targets for scattering experiments. A reader already working on NMR metrology might pick up the idea and try to reproduce it, but most others would wait for quantitative validation. It is worth sending to peer review because the underlying problem is real and the technique is new, but the referees will need to see the missing comparisons and error metrics before any stronger claim can be accepted.

Referee Report

2 major / 1 minor

Summary. The manuscript claims to present the first application of deep neural networks to continuous-wave NMR polarization metrology for solid-state dynamically polarized targets. It asserts that NN-based signal extraction and denoising achieve a substantial reduction in fitting uncertainties under realistic simulated and experimental conditions, enabling improved real-time online monitoring and higher-precision offline analysis, which in turn enhances the effective figure of merit for scattering experiments.

Significance. If the claimed uncertainty reductions are quantitatively validated and shown to generalize without new systematics, the work could provide a practical tool for reducing analysis-induced errors in polarized-target experiments, directly benefiting nuclear and high-energy physics measurements that rely on CW-NMR.

major comments (2)

[Abstract] Abstract: the central claim of 'substantial reduction of fitting uncertainties' is presented without any numerical values, error bars, baseline comparisons to conventional Q-meter methods, or specific network hyperparameters, preventing assessment of whether the improvement is load-bearing or merely incremental.
[Abstract] Abstract: the statement that performance holds 'under a variety of realistic simulated and experimental conditions' lacks any description of training/validation splits, domain-shift mitigation, or direct NN-versus-conventional uncertainty comparisons on held-out real experimental runs, leaving the generalization assumption untested in the provided text.

minor comments (1)

[Abstract] Abstract: consider adding at least one concrete quantitative result (e.g., uncertainty reduction factor or R² comparison) to allow readers to gauge the scale of the claimed improvement.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We agree that the abstract would benefit from greater quantitative specificity and have revised it to incorporate numerical results, comparisons, and methodological details drawn directly from the main text. This addresses the concerns while preserving the manuscript's core claims. We respond point by point to the major comments below.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim of 'substantial reduction of fitting uncertainties' is presented without any numerical values, error bars, baseline comparisons to conventional Q-meter methods, or specific network hyperparameters, preventing assessment of whether the improvement is load-bearing or merely incremental.

Authors: We acknowledge that the original abstract was too concise and omitted the requested quantitative elements. In the revised manuscript we have updated the abstract to include the specific uncertainty reductions, associated error bars, direct comparisons against conventional Q-meter fitting, and the network architecture plus key hyperparameters as reported in Sections 3 and 4. These additions make the magnitude of the improvement explicit and allow readers to evaluate its significance. revision: yes
Referee: [Abstract] Abstract: the statement that performance holds 'under a variety of realistic simulated and experimental conditions' lacks any description of training/validation splits, domain-shift mitigation, or direct NN-versus-conventional uncertainty comparisons on held-out real experimental runs, leaving the generalization assumption untested in the provided text.

Authors: We agree that the abstract should briefly indicate the validation strategy. We have revised it to note the use of simulated data with explicit train/validation splits, domain-adaptation steps for experimental data, and direct NN-versus-conventional comparisons performed on held-out experimental runs. These details, already present in the body of the paper, are now summarized in the abstract to support the generalization statement. revision: yes

Circularity Check

0 steps flagged

No circularity: NN application framed as external ML technique without self-referential reductions

full rationale

The paper presents the use of neural network architectures for CW-NMR signal extraction and denoising as an external machine learning application. No equations, derivations, fitted parameters, or self-citations are shown that reduce the claimed uncertainty reductions to inputs defined by the method itself. The central claim rests on empirical performance under simulated and experimental conditions rather than any self-definitional loop, fitted-input prediction, or load-bearing self-citation chain. This matches the expectation of a self-contained empirical result with no quoted reduction to its own construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that advanced ML techniques for signal extraction will outperform traditional fitting in this physics context. No free parameters, new physical entities, or ad-hoc axioms are introduced in the abstract.

axioms (1)

domain assumption Neural networks trained on simulated NMR signals can generalize to reduce uncertainties in real experimental data.
Invoked implicitly when claiming improvements under realistic simulated and experimental conditions.

pith-pipeline@v0.9.0 · 5494 in / 1179 out tokens · 52065 ms · 2026-05-15T12:46:45.395400+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 9 internal anchors

[1]

The Principles of Nuclear Magnetism

A. Abragam and L. C. Hebel. “The Principles of Nuclear Magnetism”. In:American Journal of Physics29 (12 1961).ISSN: 0002-9505.DOI: 10.1119/1.1937646. 24

work page doi:10.1119/1.1937646 1961
[2]

Anatole Abragam.The principles of nuclear magnetism. Repr. The international series of monographs on physics. Oxford: Clarendon Pr,

work page
[3]

599 pp.ISBN: 9780198512363

work page
[4]

Takuya Akiba et al.Optuna: A Next-generation Hyperparameter Optimization Framework. 2019. arXiv:1907 . 10902 [cs.LG].URL: https://arxiv.org/abs/1907.10902

work page internal anchor Pith review Pith/arXiv arXiv 2019
[5]

Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges

Michael M. Bronstein et al.Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges. 2021. arXiv:2104.13478 [cs.LG].URL: https://arxiv.org/abs/2104.13478

work page internal anchor Pith review Pith/arXiv arXiv 2021
[6]

Manipulation of spin-1 solid-state targets

Joseph Clement and Dustin Keller. “Manipulation of spin-1 solid-state targets”. In:Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment1050 (2023), p. 168177. ISSN: 0168-9002.DOI:https : / / doi . org / 10 . 1016 / j . nima . 2023 . 168177. URL:https : / / www . sciencedirect . com ...

work page 2023
[7]

A high precision Q-meter for the measurement of proton polarization in polarised targets

G R Court et al. “A high precision Q-meter for the measurement of proton polarization in polarised targets”. In:Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment324 (3 1993), pp. 433–440.ISSN: 0168-9002.DOI:https : //doi.org/10.1016/0168-9002(93)91047-Q. URL:https : / / www . sci...

work page doi:10.1016/0168-9002(93)91047-q 1993
[8]

Modeling Non-Constant Current Effects for a Series Tune NMR Q-Meter Used for Nucleon Polarization Measurements

G.R. Court and M.A. Houlden. “Modeling Non-Constant Current Effects for a Series Tune NMR Q-Meter Used for Nucleon Polarization Measurements”. In:Proceedings of the Workshop on NMR in Polarized Targets. University of Virginia, Charlottesville, V A, U.S.A., 1998, pp. 35–45

work page 1998
[9]

Solid polarized targets for nuclear and particle physics experiments

D. G. Crabb and W. Meyer. “Solid polarized targets for nuclear and particle physics experiments”. In: Annual Review of Nuclear and Particle Science47 (1997), pp. 67–109.DOI: 10.1146/annurev.nucl.47.1.67

work page doi:10.1146/annurev.nucl.47.1.67 1997
[10]

A line-shape analysis for spin-1 NMR signals

C. Dulya et al. “A line-shape analysis for spin-1 NMR signals”. In:Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment398 (2-3 1997).ISSN: 01689002.DOI: 10.1016/S0168-9002(97)00317-3

work page doi:10.1016/s0168-9002(97)00317-3 1997
[11]

Book in preparation for MIT Press

Ian Goodfellow, Yoshua Bengio, and Aaron Courville.Deep Learning. Book in preparation for MIT Press. MIT Press, 2016.URL: http://www.deeplearningbook.org

work page 2016
[12]

Springer Series in Statistics

Trevor Hastie, Robert Tibshirani, and Jerome Friedman.The Elements of Statistical Learning. Springer Series in Statistics. New York, NY , USA: Springer New York Inc., 2001

work page 2001
[13]

Kaiming He et al.Deep Residual Learning for Image Recognition. 2015. arXiv:1512 . 03385 [cs.CV]. URL:https://arxiv.org/abs/1512.03385

work page internal anchor Pith review Pith/arXiv arXiv 2015
[14]

Multilayer feedforward networks are universal approximators

Kurt Hornik, Maxwell Stinchcombe, and Halbert White. “Multilayer feedforward networks are universal approximators”. In:Neural Networks 2.5 (1989), pp. 359–366.ISSN: 0893-6080.DOI: https : / / doi . org / 10 . 1016 / 0893 - 6080(89 ) 90020 - 8.URL: https://www.sciencedirect.com/science/ article/pii/0893608089900208

work page arXiv 1989
[15]

Jie Hu et al.Squeeze-and-Excitation Networks

work page
[16]

Squeeze-and-Excitation Networks

arXiv:1709 . 01507 [cs.CV].URL: https://arxiv.org/abs/1709.01507

work page internal anchor Pith review Pith/arXiv arXiv
[17]

Modeling alignment enhancement for solid polarized targets

D. Keller. “Modeling alignment enhancement for solid polarized targets”. In:European Physical Journal A53 (7 2017).ISSN: 1434601X.DOI: 10.1140/epja/i2017-12344-0

work page doi:10.1140/epja/i2017-12344-0 2017
[18]

Uncertainty minimization in NMR measurements of dynamic nuclear polarization of a proton target for nuclear physics experiments

D. Keller. “Uncertainty minimization in NMR measurements of dynamic nuclear polarization of a proton target for nuclear physics experiments”. In: Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment728 (2013), pp. 133–144.ISSN: 0168-9002.DOI:https : / / doi . org / 10 . 1016 / j . n...

work page 2013
[19]

Enhanced tensor polarization in solid-state targets

D. Keller, D. Crabb, and D. Day. “Enhanced tensor polarization in solid-state targets”. In:Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment981 (2020).ISSN: 01689002.DOI:10.1016/j.nima.2020.164504

work page doi:10.1016/j.nima.2020.164504 2020
[20]

A technique for Measurement of Vector and Tensor Polarization in Solid Spin One Polarized Targets

William Frederick Kielhorn. “A technique for Measurement of Vector and Tensor Polarization in Solid Spin One Polarized Targets”. In:Proceedings of the 4th International Workshop on Polarized Target Materials and Techniques. 1984

work page 1984
[21]

Teerath Kumar et al.Image Data Augmentation Approaches: A Comprehensive Survey and Future directions. 2023. arXiv:2301 . 02830 [cs.CV]. URL:https://arxiv.org/abs/2301.02830

work page arXiv 2023
[22]

Convolutional networks and applications in vision

Yann LeCun, Koray Kavukcuoglu, and Cl´ement Farabet. “Convolutional networks and applications in vision”. In:Circuits and Systems 25 (ISCAS), Proceedings of 2010 IEEE International Symposium on. IEEE. 2010, pp. 253–256

work page 2010
[23]

NN-SVG: Publication-Ready Neural Network Architecture Schematics

Alexander LeNail. “NN-SVG: Publication-Ready Neural Network Architecture Schematics”. In: Journal of Open Source Software4.33 (2019), p. 747.DOI:10 . 21105 / joss . 00747.URL: https://doi.org/10.21105/joss.00747

work page doi:10.21105/joss.00747 2019
[24]

Ilya Loshchilov and Frank Hutter.Decoupled Weight Decay Regularization. 2019. arXiv: 1711 . 05101 [cs.LG].URL: https://arxiv.org/abs/1711.05101

work page internal anchor Pith review Pith/arXiv arXiv 2019
[25]

Ilya Loshchilov and Frank Hutter.SGDR: Stochastic Gradient Descent with Warm Restarts. 2017. arXiv: 1608 . 03983 [cs.LG].URL:https : / / arxiv . org/abs/1608.03983

work page internal anchor Pith review Pith/arXiv arXiv 2017
[26]

Tensor Polarized Deuteron Target for Intermediate Energy Physics

W. Meyer and E. Schilling. “Tensor Polarized Deuteron Target for Intermediate Energy Physics”. In:Proceedings of the 4th International workshop on Polarized Target Materials and Techniques. 1984

work page 1984
[27]

Umberto Michelucci.An Introduction to Autoencoders. 2022. arXiv:2201.03898 [cs.LG]. URL:https://arxiv.org/abs/2201.03898

work page arXiv 2022
[28]

Adam Paszke et al.PyTorch: An Imperative Style, High-Performance Deep Learning Library. 2019. arXiv:1912 . 01703 [cs.LG].URL: https://arxiv.org/abs/1912.01703

work page internal anchor Pith review Pith/arXiv arXiv 2019
[29]

Workshop on NMR in Polarization Targets

S. I. Pentill ¨a. “Workshop on NMR in Polarization Targets”. In:Proceedings of the Workshop on NMR in Polarized Targets. Charlottesville, V A: University of Virginia, 1998, pp. 15–26

work page 1998
[30]

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

M. Raissi, P. Perdikaris, and G.E. Karniadakis. “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations”. In:Journal of Computational Physics 378 (2019), pp. 686–707.ISSN: 0021-9991.DOI: https : / / doi . org / 10 . 1016 / j . jcp . 2018 . 10 . 045. URL:https ...

work page 2019
[31]

A survey on image data augmentation for deep learning

Connor Shorten and Taghi M Khoshgoftaar. “A survey on image data augmentation for deep learning”. In:Journal of Big Data6.1 (2019), pp. 1–48

work page 2019
[32]

Stolte et al.DOMINO++: Domain-aware Loss Regularization for Deep Learning Generalizability

Skylar E. Stolte et al.DOMINO++: Domain-aware Loss Regularization for Deep Learning Generalizability. 2023. arXiv:2308 . 10453 [cs.CV].URL: https://arxiv.org/abs/2308.10453

work page arXiv 2023
[33]

Christian Szegedy et al.Going Deeper with Convolutions. 2014. arXiv:1409 . 4842 [cs.CV]. URL:https://arxiv.org/abs/1409.4842

work page internal anchor Pith review Pith/arXiv arXiv 2014
[34]

New York: Wiley, 1998.ISBN: 978-0-471-03003-4

Vladimir N Vapnik.Statistical Learning Theory. New York: Wiley, 1998.ISBN: 978-0-471-03003-4

work page 1998
[35]

Ashish Vaswani et al.Attention Is All You Need

work page
[36]

Attention Is All You Need

arXiv:1706 . 03762 [cs.CL].URL: https://arxiv.org/abs/1706.03762

work page internal anchor Pith review Pith/arXiv arXiv
[37]

English (US)

Ragav Venkatesan and Baoxin Li.Convolutional Neural Networks in Visual Computing: A Concise Guide. English (US). Publisher Copyright: © 2018 by Taylor & Francis Group, LLC. All rights reserved. CRC Press, Oct. 2017.ISBN: 9781498770392.DOI:10.4324/9781315154282

work page doi:10.4324/9781315154282 2018
[38]

Extracting and composing robust features with denoising autoencoders

Pascal Vincent et al. “Extracting and composing robust features with denoising autoencoders.” In: ICML. Ed. by William W. Cohen, Andrew McCallum, and Sam T. Roweis. V ol. 307. ACM International Conference Proceeding Series. ACM, 2008, pp. 1096–1103.ISBN: 978-1-60558-205-4.URL: http://dblp.uni- trier.de/db/conf/icml/ icml2008.html#VincentLBM08

work page 2008
[39]

Data Augmentation for Deep Learning: A Survey

Qingyun Zhang et al. “Data Augmentation for Deep Learning: A Survey”. In:Journal of Big Data8.1 (2021), pp. 1–48.DOI:10 . 1186 / s40537 - 021 - 00499-4

work page 2021