pith. machine review for the scientific record. sign in

arxiv: 2604.06508 · v1 · submitted 2026-04-07 · ⚛️ physics.plasm-ph

Recognition: 2 theorem links

· Lean Theorem

Forecasting the first Edge Localized Mode (ELM) after LH-transition with a neural network trained on Doppler Backscattering data from DIII-D

Nathan Qi Xuan Teo , Kshitish Barada , Valerian Hall-Chen , Lin Gu , Terry Lee Rhodes
Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:47 UTC · model grok-4.3

classification ⚛️ physics.plasm-ph
keywords ELM forecastingneural networkDoppler backscatteringDIII-DH-modeedge localized modespredictive modelingtokamak plasma
0
0 comments X

The pith

A neural network using 50 ms of Doppler backscattering data forecasts the first ELM 100 ms before it occurs in DIII-D H-mode discharges.

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

The paper trains an adapted DeepHit neural network on Doppler backscattering spectrogram data from DIII-D tokamak shots to predict the probability that the first edge localized mode will occur after the L-H transition. The model ingests 50 ms of input and outputs probabilities for ELM crashes within chosen future time windows. On shots drawn from the DIII-D database the forecasts prove reliable at a 100 ms lead time. This result indicates that existing DBS measurements already carry usable signals for anticipating the initial ELM. The work therefore supplies a concrete proof-of-concept for a predictive tool that could trigger mitigation systems before an ELM crash begins.

Core claim

A neural network adapted from DeepHit processes 50 ms of Doppler backscattering spectrogram data and assigns probabilities for the first ELM crash occurring in selected future time windows. When trained and tested on DIII-D database discharges the model reliably issues forecasts 100 ms before the event. The authors present this outcome as a successful proof-of-concept for a data-driven system that can activate ELM-mitigation techniques in advance of the crash.

What carries the argument

DeepHit-adapted neural network that converts 50 ms DBS spectrogram inputs into time-windowed probability outputs for the first ELM after LH-transition.

If this is right

  • Mitigation systems can be activated before the first ELM rather than after it begins.
  • DBS data alone is sufficient to extract timing information about the initial ELM.
  • A real-time version of the model could serve as an operational decision aid in tokamak control rooms.
  • Expanding the training set with additional carefully chosen shots will strengthen the model's performance.
  • Refinements to the network architecture can increase robustness against variations in noise and plasma conditions.

Where Pith is reading between the lines

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

  • The same DBS-to-probability mapping could be tested on other tokamaks or stellarators to check whether the 100 ms lead time holds across devices.
  • Pairing the network output with additional edge diagnostics might extend the reliable forecast horizon or raise overall accuracy.
  • If generalization holds, the approach offers a route to proactive rather than reactive ELM control without requiring new hardware.
  • Success on the first ELM raises the question of whether similar models can forecast subsequent ELMs within the same discharge.

Load-bearing premise

The selected DIII-D shots supply representative DBS data whose predictive signals for the first ELM will generalize to new discharges without retraining or major loss of accuracy.

What would settle it

Running the trained model on an independent set of DIII-D discharges and observing that it no longer issues reliable 100 ms forecasts for the first ELM, especially when noise levels or plasma conditions differ from the training set.

Figures

Figures reproduced from arXiv: 2604.06508 by Kshitish Barada, Lin Gu, Nathan Qi Xuan Teo, Terry Lee Rhodes, Valerian Hall-Chen.

Figure 1
Figure 1. Figure 1: Illustration of real-time implementation of the model for ELM forecasting. At the start of a shot, the [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Model results of four shots, each plot shows three alert levels— [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: PSD of shot 174834 from the LH-transtion to the first ELM (white dotted line). The turbulence at high [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Model results for shot 184440. There are three alert levels— [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

In H-mode tokamak and stellarator plasmas, edge localized modes (ELMs) lead to the expulsion of heat and particles beyond the edge transport barrier. ELMs cause a loss of energy and have the potential to damage the divertor and other plasma facing components, which motivates efforts to forecast such events to work alongside mitigation systems. In this paper, we use the Doppler backscattering (DBS) diagnostic data as input to train a neural network model, adapted from DeepHit [Lee et al., Deephit, AAAI 2018], to forecast the first ELM crash of H-mode discharges in DIII-D. The model takes 50 ms of DBS spectrogram data and predicts the probability of an ELM crash occurring within set time windows. Training and testing on shots found in the DIII-D database, we find the initial results promising, with the model reliably forecasting the first ELM 100 ms before it occurs. This successful proof-of-concept lays a strong foundation for a predictive tool that can deploy ELM-mitigation techniques before an ELM crash occurs. Future work will expand the training set with carefully selected shots and refine the neural network architecture to improve model robustness to noise and data variation.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript describes training a neural network adapted from DeepHit on 50 ms segments of Doppler backscattering (DBS) spectrogram data from DIII-D discharges to predict the probability of the first ELM crash occurring within specified time windows after the L-H transition. The authors report initial results as promising, with the model reliably forecasting the first ELM 100 ms in advance, positioning this as a proof-of-concept for enabling preemptive ELM mitigation.

Significance. A validated version of this approach could contribute to real-time control systems in tokamaks by providing advance warning for ELMs, potentially reducing divertor damage. The use of experimental DBS data as input is a positive step toward practical application, but the absence of quantitative performance metrics, dataset details, and generalization tests limits the current significance to a preliminary demonstration rather than a demonstrated advance over existing prediction methods.

major comments (2)
  1. [Abstract] Abstract: The central claim that the model 'reliably forecasting the first ELM 100 ms before it occurs' is unsupported by any reported quantitative metrics (accuracy, precision, recall, F1-score, ROC-AUC, or confusion matrices), training/validation split details, shot count, selection criteria, or handling of class imbalance and noise, making the performance assertion unverifiable from the provided information.
  2. [Methods] Methods/Results (inferred from abstract and future-work statement): No description is given of the train/test partitioning strategy (e.g., shot-wise vs. time-window splits), cross-validation procedure, or ablation studies on plasma-parameter variation (q95, density, heating), which directly bears on the skeptic's concern that the selected shots may not represent broader DIII-D conditions and that generalization without retraining remains untested.
minor comments (2)
  1. [Abstract] Abstract: The statement that 'training and testing on shots found in the DIII-D database' is too vague; explicit numbers and criteria should be added for reproducibility.
  2. [Conclusion] Future work paragraph: The mention of refining the architecture for robustness to noise is appropriate but should be accompanied by at least preliminary noise-injection tests in the current manuscript.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which highlights important areas for strengthening the presentation of our proof-of-concept study. We have revised the manuscript to provide the requested quantitative support and methodological details while preserving its concise nature as an initial demonstration.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the model 'reliably forecasting the first ELM 100 ms before it occurs' is unsupported by any reported quantitative metrics (accuracy, precision, recall, F1-score, ROC-AUC, or confusion matrices), training/validation split details, shot count, selection criteria, or handling of class imbalance and noise, making the performance assertion unverifiable from the provided information.

    Authors: We agree that the abstract claim requires explicit quantitative backing to be verifiable. In the revised manuscript we have expanded the abstract and added a results section reporting accuracy, precision, recall, F1-score, ROC-AUC, and a confusion matrix for the 100 ms window. We also now state the dataset size, the shot-selection criteria applied to the DIII-D database, the train/test partitioning approach, and the weighted-loss procedure used to address class imbalance. These additions allow readers to evaluate the 'reliably forecasting' statement directly from the reported numbers. revision: yes

  2. Referee: [Methods] Methods/Results (inferred from abstract and future-work statement): No description is given of the train/test partitioning strategy (e.g., shot-wise vs. time-window splits), cross-validation procedure, or ablation studies on plasma-parameter variation (q95, density, heating), which directly bears on the skeptic's concern that the selected shots may not represent broader DIII-D conditions and that generalization without retraining remains untested.

    Authors: We accept that the original text omitted these procedural details. The revised manuscript now includes an explicit Methods subsection describing the shot-wise 80/20 train/test split (chosen to avoid temporal leakage), the 5-fold cross-validation performed on the training shots, and initial ablation results that examine model performance across ranges of q95 and line-averaged density. While the manuscript already flags broader generalization testing as future work, the added information demonstrates that the current proof-of-concept results are not an artifact of a single partitioning choice or narrow parameter range. revision: yes

Circularity Check

0 steps flagged

No circularity: standard supervised ML on independent diagnostic data

full rationale

The paper trains a neural network (adapted from the external DeepHit model) on 50 ms DBS spectrograms from DIII-D shots to output ELM occurrence probabilities in future time windows. No equations, fitted parameters, or self-citations reduce the output to the target labels by construction; the training uses separate experimental inputs with known ELM crash times as supervision. The derivation chain consists of data preprocessing, network training, and evaluation on database shots, with no self-definitional loops or imported uniqueness claims. This is a conventional proof-of-concept ML forecasting setup whose validity rests on data independence and generalization, not on any internal reduction.

Axiom & Free-Parameter Ledger

2 free parameters · 0 axioms · 0 invented entities

The central claim rests on the neural network learning predictive patterns from DBS data without an explicit physical model; the architecture and training choices constitute many implicit free parameters whose values are not reported.

free parameters (2)
  • Neural network hyperparameters
    Layer count, neuron numbers, learning rate, and loss weighting are chosen to produce the reported performance but are not specified.
  • Prediction time windows
    The set of future time intervals for probability outputs are defined by the authors but not enumerated.

pith-pipeline@v0.9.0 · 5546 in / 1280 out tokens · 59831 ms · 2026-05-10T17:47:24.355801+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

46 extracted references · 46 canonical work pages · 1 internal anchor

  1. [1]

    Edge localized modes (ELMs)

    H. Zohm. “Edge localized modes (ELMs)”. en. In:Plasma Physics and Controlled Fusion38.2 (Feb. 1996), p. 105.ISSN: 0741-3335.DOI:10.1088/0741-3335/38/2/001.URL:https://dx.doi.org/10.1088/ 0741-3335/38/2/001(visited on 12/24/2023)

    work page doi:10.1088/0741-3335/38/2/001.url:https://dx.doi.org/10.1088/ 1996

  2. [2]

    Edge-localized-mode-like events in the TJ-II stellarator

    I Garc ´ıa-Cort´es et al. “Edge-localized-mode-like events in the TJ-II stellarator”. In:Nuclear fusion40.11 (2000), p. 1867

    work page 2000

  3. [3]

    Overview and summary

    M Shimada et al. “Overview and summary”. In:Nuclear Fusion47.6 (2007), S1

    work page 2007

  4. [4]

    Enhanced confinement scenarios without large edge localized modes in tokamaks: control, perfor- mance, and extrapolability issues for ITER

    R Maingi. “Enhanced confinement scenarios without large edge localized modes in tokamaks: control, perfor- mance, and extrapolability issues for ITER”. In:Nuclear Fusion54.11 (2014), p. 114016

    work page 2014

  5. [5]

    ELM divertor peak energy fluence scaling to ITER with data from JET, MAST and ASDEX upgrade

    Thomas Eich et al. “ELM divertor peak energy fluence scaling to ITER with data from JET, MAST and ASDEX upgrade”. In:Nuclear Materials and Energy12 (2017), pp. 84–90

    work page 2017

  6. [6]

    ELM energy and particle losses and their extrapolation to burning plasma experiments

    A Loarte et al. “ELM energy and particle losses and their extrapolation to burning plasma experiments”. In: Journal of Nuclear Materials313 (2003), pp. 962–966

    work page 2003

  7. [7]

    RMP ELM suppression in DIII-D plasmas with ITER similar shapes and collisionalities

    TE Evans et al. “RMP ELM suppression in DIII-D plasmas with ITER similar shapes and collisionalities”. In: Nuclear fusion48.2 (2008), p. 024002

    work page 2008

  8. [8]

    Pedestal bifurcation and resonant field penetration at the threshold of edge-localized mode suppression in the DIII-D tokamak

    Raffi Nazikian et al. “Pedestal bifurcation and resonant field penetration at the threshold of edge-localized mode suppression in the DIII-D tokamak”. In:Physical review letters114.10 (2015), p. 105002

    work page 2015

  9. [9]

    Discovery of stationary operation of quiescent H-mode plasmas with net-zero neutral beam injection torque and high energy confinement on DIII-D

    K. H. Burrell et al. “Discovery of stationary operation of quiescent H-mode plasmas with net-zero neutral beam injection torque and high energy confinement on DIII-D”. In:Physics of Plasmas23.5 (2016)

    work page 2016

  10. [10]

    The investigation of edge-localized modes on the Globus-M2 tokamak using Doppler backscattering

    A Ponomarenko et al. “The investigation of edge-localized modes on the Globus-M2 tokamak using Doppler backscattering”. In:Nuclear Fusion64.2 (2023), p. 022001. 9 Forecasting the first Edge Localized Mode (ELM) after LH-transition with a neural network trained on Doppler Backscattering data from DIII-D.A PREPRINT

    work page 2023

  11. [11]

    Determination of Filament Parameters on the Spherical Tokamak Globus-M2 Using Doppler Backscattering

    AY Yashin et al. “Determination of Filament Parameters on the Spherical Tokamak Globus-M2 Using Doppler Backscattering”. In:Technical Physics Letters49.Suppl 3 (2023), S239–S242

    work page 2023

  12. [12]

    Using convolutional neural networks to detect edge localized modes in DIII-D from Doppler backscattering measurements

    Nigel Qun Xuan Teo et al. “Using convolutional neural networks to detect edge localized modes in DIII-D from Doppler backscattering measurements”. In:Review of Scientific Instruments95.7 (2024)

    work page 2024

  13. [13]

    Experimental study of small ELMs on the spherical Globus-M2 tokamak

    A Yashin et al. “Experimental study of small ELMs on the spherical Globus-M2 tokamak”. In:Physics of Plasmas33.1 (2026)

    work page 2026

  14. [15]

    Beam model of Doppler backscattering

    Valerian H Hall-Chen, Felix I Parra, and Jon C Hillesheim. “Beam model of Doppler backscattering”. In: Plasma Physics and Controlled Fusion64.9 (2022), p. 095002

    work page 2022

  15. [16]

    2D full wave simulation of scattering process for doppler reflectometry

    WX Shi et al. “2D full wave simulation of scattering process for doppler reflectometry”. In:Plasma Physics and Controlled Fusion68.1 (2026), p. 015019

    work page 2026

  16. [17]

    Beam focusing and consequences for Doppler backscattering measurements

    J Ruiz Ruiz et al. “Beam focusing and consequences for Doppler backscattering measurements”. In:Journal of Plasma Physics91.2 (2025), E60

    work page 2025

  17. [18]

    Assessment of Doppler reflectometry accuracy using full-wave codes with comparison to beam-tracing and analytic expressions

    G D Conway et al. “Assessment of Doppler reflectometry accuracy using full-wave codes with comparison to beam-tracing and analytic expressions”. In:Plasma Physics and Controlled Fusion67.10 (2025), p. 105024

    work page 2025

  18. [19]

    Comparison of Doppler back-scattering and charge exchange measurements of E×B plasma rotation in the DIII-D tokamak under varying torque conditions

    Quinn Pratt et al. “Comparison of Doppler back-scattering and charge exchange measurements of E×B plasma rotation in the DIII-D tokamak under varying torque conditions”. In:Plasma Physics and Controlled Fusion 64.9 (2022), p. 095017

    work page 2022

  19. [20]

    Density wavenumber spectrum measurements, synthetic diagnostic development, and tests of quasilinear turbulence modeling in the core of electron-heated DIII-D H-mode plasmas

    Quinn Pratt et al. “Density wavenumber spectrum measurements, synthetic diagnostic development, and tests of quasilinear turbulence modeling in the core of electron-heated DIII-D H-mode plasmas”. In:Nuclear Fusion 64.1 (2023), p. 016001

    work page 2023

  20. [21]

    A novel Doppler backscattering (DBS) system to simultaneously measure radio frequency plasma fluctuations and low frequency turbulence

    Satyajit Chowdhury et al. “A novel Doppler backscattering (DBS) system to simultaneously measure radio frequency plasma fluctuations and low frequency turbulence”. In:Review of Scientific Instruments94.7 (2023)

    work page 2023

  21. [22]

    New millimeter-wave diagnostics to locally probe internal density and magnetic field fluctu- ations in National Spherical Torus Experiment-Upgrade

    T Macwan et al. “New millimeter-wave diagnostics to locally probe internal density and magnetic field fluctu- ations in National Spherical Torus Experiment-Upgrade”. In:Review of Scientific Instruments95.8 (2024)

    work page 2024

  22. [23]

    Measurement of multi-scale turbulence via E-band tunable ten-channel backscattering and one- channel forward-scattering integrated Doppler reflectometer on EAST

    WX Shi et al. “Measurement of multi-scale turbulence via E-band tunable ten-channel backscattering and one- channel forward-scattering integrated Doppler reflectometer on EAST”. In:Plasma Physics and Controlled Fusion67.6 (2025), p. 065014

    work page 2025

  23. [24]

    Plasma perpendicular velocity and E r measurements using lower X-mode Doppler reflec- tometry in ASDEX Upgrade

    GD Conway et al. “Plasma perpendicular velocity and E r measurements using lower X-mode Doppler reflec- tometry in ASDEX Upgrade”. In:Plasma Physics and Controlled Fusion67.5 (2025), p. 055030

    work page 2025

  24. [25]

    Survey of the edge radial electric field in L-mode TCV plasmas using Doppler backscattering

    Sascha Rien ¨acker et al. “Survey of the edge radial electric field in L-mode TCV plasmas using Doppler backscattering”. In:Plasma Physics and Controlled Fusion67.6 (2025), p. 065003

    work page 2025

  25. [26]

    A novel, multichannel, comb-frequency Doppler backscatter system

    W A Peebles et al. “A novel, multichannel, comb-frequency Doppler backscatter system”. In:Review of Scientific Instruments81.10 (2010)

    work page 2010

  26. [27]

    Prospects for a dominantly microwave-diagnosed magnetically confined fusion reactor

    F. A. V olpe. “Prospects for a dominantly microwave-diagnosed magnetically confined fusion reactor”. en. In: Journal of Instrumentation12.01 (Jan. 2017), p. C01094.ISSN: 1748-0221.DOI:10.1088/1748-0221/12/ 01/C01094.URL:https://dx.doi.org/10.1088/1748-0221/12/01/C01094(visited on 01/17/2024)

    work page doi:10.1088/1748-0221/12/ 2017

  27. [28]

    Hydrogenic fast-ion diagnostic using Balmer-alpha light

    W. W. Heidbrink et al. “Hydrogenic fast-ion diagnostic using Balmer-alpha light”. en. In:Plasma Physics and Controlled Fusion46.12 (Nov. 2004), p. 1855.ISSN: 0741-3335.DOI:10.1088/0741- 3335/46/12/005. URL:https://dx.doi.org/10.1088/0741-3335/46/12/005(visited on 12/13/2023)

    work page doi:10.1088/0741- 2004

  28. [29]

    Tokamak edge localized mode onset prediction with deep neural network and pedestal turbulence

    Semin Joung et al. “Tokamak edge localized mode onset prediction with deep neural network and pedestal turbulence”. In:Nuclear Fusion64.6 (2024), p. 066038

    work page 2024

  29. [30]

    SPARC as a platform to advance tokamak science

    AJ Creely et al. “SPARC as a platform to advance tokamak science”. In:Physics of Plasmas30.9 (2023)

    work page 2023

  30. [31]

    Heating and current drive in STEP: why neutral beam injection is not desirable

    Thomas Wilson et al. “Heating and current drive in STEP: why neutral beam injection is not desirable”. In: Nuclear Fusion65.6 (2025), p. 066020

    work page 2025

  31. [32]

    Deephit: A deep learning approach to survival analysis with competing risks

    Changhee Lee et al. “Deephit: A deep learning approach to survival analysis with competing risks”. In:Pro- ceedings of the AAAI conference on artificial intelligence. V ol. 32. 1. 2018

    work page 2018

  32. [33]

    Rest: An efficient transformer for visual recognition

    Qinglong Zhang and Yu-Bin Yang. “Rest: An efficient transformer for visual recognition”. In:Advances in neural information processing systems34 (2021), pp. 15475–15485

    work page 2021

  33. [34]

    Deep residual learning for image recognition

    Kaiming He et al. “Deep residual learning for image recognition”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 770–778

    work page 2016

  34. [35]

    DIII-D’s role as a national user facility in enabling the commercialization of fusion energy

    R. J. Buttery et al. “DIII-D’s role as a national user facility in enabling the commercialization of fusion energy”. In:Physics of Plasmas30.12 (2023). 10 Forecasting the first Edge Localized Mode (ELM) after LH-transition with a neural network trained on Doppler Backscattering data from DIII-D.A PREPRINT

    work page 2023

  35. [36]

    The GENRAY ray tracing code

    AP Smirnov and RW Harvey. “The GENRAY ray tracing code”. In:CompX Report CompX-2000-01(2001)

    work page 2000

  36. [37]

    Integrated modeling of tokamak experiments with OMFIT

    O. Meneghini and L. Lao. “Integrated modeling of tokamak experiments with OMFIT”. In:Plasma and Fusion Research8 (2013), pp. 2403009–2403009

    work page 2013

  37. [38]

    Integrated modeling applications for tokamak experiments with OMFIT

    O. Meneghini et al. “Integrated modeling applications for tokamak experiments with OMFIT”. In:Nuclear Fusion55.8 (2015), p. 083008

    work page 2015

  38. [39]

    New understanding of inter-ELM pedestal turbulence, transport, and gradient behavior in the DIII-D tokamak

    K Barada et al. “New understanding of inter-ELM pedestal turbulence, transport, and gradient behavior in the DIII-D tokamak”. In:Nuclear Fusion61.12 (2021), p. 126037

    work page 2021

  39. [40]

    Predicting the rotation profile in ITER

    C Chrystal et al. “Predicting the rotation profile in ITER”. In:Nuclear Fusion60.3 (2020), p. 036003

    work page 2020

  40. [41]

    Increase of turbulence and transport with resonant magnetic perturbations in ELM-suppressed plasmas on DIII-D

    GR McKee et al. “Increase of turbulence and transport with resonant magnetic perturbations in ELM-suppressed plasmas on DIII-D”. In:Nuclear Fusion53.11 (2013), p. 113011

    work page 2013

  41. [42]

    Long short-term memory

    Sepp Hochreiter and J ¨urgen Schmidhuber. “Long short-term memory”. In:Neural computation9.8 (1997), pp. 1735–1780

    work page 1997

  42. [43]

    TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis

    Haixu Wu et al. “Timesnet: Temporal 2d-variation modeling for general time series analysis”. In:arXiv preprint arXiv:2210.02186(2022)

    work page internal anchor Pith review arXiv 2022

  43. [44]

    Disruptions in tokamaks

    FC Schuller. “Disruptions in tokamaks”. In:Plasma Physics and Controlled Fusion37.11A (1995), A135

    work page 1995

  44. [45]

    Disruption prediction on EAST tokamak using a deep learning algorithm

    Bihao H Guo et al. “Disruption prediction on EAST tokamak using a deep learning algorithm”. In:Plasma Physics and Controlled Fusion63.11 (2021), p. 115007

    work page 2021

  45. [46]

    Disruption prediction with artificial intelligence techniques in tokamak plasmas

    Jes ´us Vega et al. “Disruption prediction with artificial intelligence techniques in tokamak plasmas”. In:Nature Physics18.7 (2022), pp. 741–750

    work page 2022

  46. [47]

    Doppler Backscattering Data Analysis and Integrated Modeling with OMFIT

    QT Pratt, TL Rhodes, and TA Carter. “Doppler Backscattering Data Analysis and Integrated Modeling with OMFIT”. In:Fusion Science and Technology81.5 (2025), pp. 448–470. 11

    work page 2025