Pith. sign in

REVIEW 3 major objections 6 minor 100 references

Plasma State Monitoring and Disruption Characterization using Multimodal VAEs

T0 review · 3 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A multimodal VAE maps TCV discharges into a latent space where a calibrated disruption risk tracks the empirical disruption rate.

desk verdict Useful methods paper with a convincing latent-space separation of disruption types, but the quantitative risk calibration and the DEFUSE tD labels carry more weight than they should. read the letter →

arxiv 2504.17710 v1 pith:MWVK5YUO submitted 2025-04-24 physics.plasm-ph cs.CVcs.LG

Yoeri Poels , Alessandro Pau , Christian Donner , Giulio Romanelli , Olivier Sauter , Cristina Venturini , Vlado Menkovski , the TCV team
show 1 more author
the WPTE team
This is my paper · ORCID
classification physics.plasm-phcs.CVcs.LG
keywords plasmadisruptiontokamakvariationalautoencoderlatentvariablemodelriskinterpretablemachinelearningTCVcounterfactualanalysis
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 a multimodal variational autoencoder can turn tokamak discharge measurements into a low-dimensional, interpretable map of plasma operating regimes, with a calibrated disruption-risk coordinate that tracks how likely a discharge is to disrupt. The authors argue that such a representation complements black-box disruption predictors by making the proximity to disruption visible and separable across different disruption types. They demonstrate the claim on roughly 1,600 TCV flat-top discharges, showing that the learned risk variable matches the empirical disruption rate and that ITER baseline, density limit, and negative triangularity disruptions end in distinct latent states. A sympathetic reader would take the central claim to be that data-driven plasma-state monitoring can be both expressive and interpretable enough to support disruption characterization and downstream causal-style analyses.

What carries the argument

The central object is a two-dimensional latent state $z$ whose trajectory represents the plasma state over time, together with a learned disruption-risk map $D_{\text{risk}}(z)\in[0,1]$. The encoder is a residual sequential model that updates the latent mean from a sliding window of diagnostic signals, the prior is a Gaussian mixture with $K=8$ modes that encourages multimodal clustering, and the decoder maps each latent point back to the full signal vector so the space can be read off in physics quantities. The disruption-risk head is trained with binary cross-entropy labels that ramp to 1 during the second before the disruption time $t_D$, and the final risk values are recalibrated with Platt scaling; this risk term is what pulls disruptive and non-disruptive regions apart in the latent space.

What would settle it

Take a held-out subset of TCV discharges and manually label the disruption onset time (start of the thermal quench) without reference to the automated detections, then retrain or re-evaluate the risk calibration with those labels; if the expected calibration error rises well above the reported test value or the three disruption families no longer fall into distinct terminal states, the central claim would be refuted.

Watch

Extended reading notes

Core claim

The central discovery is that a two-dimensional latent variable, learned by a sequential multimodal VAE with an added disruption-risk head, organizes the TCV flat-top operational space into smooth regions whose calibrated disruption risk $D_{\text{risk}}$ closely follows the empirically observed disruption rate. After post-hoc Platt scaling, the expected calibration error is 0.0307 on the training set and 0.0715 on a held-out test set, and low-risk regions contain almost no disrupting discharges while risk rises exponentially near disruption. The same latent space separates known disruption families: projections of ITER baseline, density limit, and negative triangularity discharges start clustered together but terminate in different peaks of high-risk regions, and component planes show physically sensible correlations with $q_{95}$, Greenwald fraction, and vertical growth rate. The method also supports a proof-of-principle counterfactual analysis in which the most similar non-disrupting counterpart of a disrupting shot is found automatically, and the most distinct signal differences recover known disruption drivers such as MHD activity, density-limit proximity, and vertical instability.

Load-bearing premise

The evaluation depends on the automated disruption-onset times, event detections, and confinement-state labels being accurate enough to define both the training targets and the evaluation metadata; if those onset times are systematically wrong, the risk calibration and the terminal-state clustering would shift even if the VAE itself is sound.

Editorial extensions

If this is right

  • The calibrated $D_{\text{risk}}$ variable can serve as a continuous, interpretable indicator of disruption rate across the flat-top operational space, with roughly 3% average calibration error on the training set and 7% on novel discharges.
  • The method automatically separates at least three distinct disruption families into different terminal latent states, meaning it can be used to label the type of disruption that a trajectory is heading toward.
  • Latent states correlate with plasma confinement states even though no confinement labels were used in training, so the learned map recovers physically meaningful operating regimes without supervision.
  • The counterfactual-style analysis identifies known disruption-related parameters for different scenarios, suggesting the latent space can automate the search for differences between disrupting and non-disrupting shots.
  • Because the decoder projects each latent point back to physics quantities, the same map can be reused to inspect which measured signals change as a discharge moves toward higher-risk regions.

Reading between the lines

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

  • Beyond the paper, the same latent-map construction could be extended to multi-machine databases, although device-specific time and spatial scales would need to be handled explicitly rather than letting each machine collapse into its own latent region.
  • The 5 ms latent timestep and slow-timescale focus mean the method is better suited to identifying global regime proximity than to resolving the fast chain-of-events immediately before a disruption; a dedicated early-warning comparison against existing predictors would be a natural next test.
  • If $D_{\text{risk}}$ remains well calibrated under online projection of new discharges, it could be embedded in real-time control as a risk coordinate to steer trajectories away from high-risk regions, which the paper only sketches as future work.
  • A direct falsifiable extension would be to recompute the calibration error using independently labeled disruption-onset times rather than the automated detections, since the current risk labels inherit any error in those detections.
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 / 6 minor

Summary. The paper proposes a multimodal sequential variational autoencoder (VAE) for modeling TCV flat-top plasma discharges in a 2D latent space with a Gaussian mixture prior. The model is extended with a disruption-risk head trained on binary labels derived from DEFUSE disruption onset times, and the resulting risk variable is post hoc Platt-scaled to match an empirical disruption rate. The authors evaluate the latent space qualitatively and quantitatively: they report an expected calibration error of 0.0307 on training data and 0.0715 on held-out test data, show that latent states correlate with confinement states and disruption-related events, demonstrate that ITER Baseline Scenario, density limit, and negative triangularity disruptions separate into distinct terminal latent states, and run a proof-of-principle counterfactual analysis that identifies disruption-correlated parameters consistent with known physics.

Significance. If the central claims hold, this is a valuable contribution to disruption characterization and plasma state monitoring. The method is deliberately interpretable, in contrast to most black-box disruption predictors, and the qualitative findings are physically meaningful: low q95 near IBL disruptions, high Greenwald fraction near DL disruptions, and elevated vertical growth rate near NT disruptions. The paper is also unusually transparent about its limitations, including ongoing event-detection development (Footnote 4) and manual hyperparameter selection (Appendix A). Strengths include a clearly described dataset, careful attention to causal interpolation to avoid information leakage, and downstream analysis that produces falsifiable, physics-aligned predictions. However, the quantitative calibration claim is weakened by the post hoc Platt scaling and by the shared reliance on the same DEFUSE t_D labels for training labels, calibration targets, and terminal-state assignments, which limits the strength of the evidence for the headline ECE values.

major comments (3)
  1. [Section 5.2, Eq. (37), with Eqs. (29)-(30)] The reported ECE values are computed after fitting Platt scaling on the same training set, and the empirical disruption rate eDrate used as the calibration target is defined using the same t_D that generates the y_tm training labels in Eq. (29). The training ECE of 0.0307 therefore measures how well a one-parameter calibrated curve fits the empirical rate on the data it was fit to, rather than how well D_risk predicts an independent quantity. Please report the ECE of the raw, uncalibrated D_risk before Platt scaling, and use a proper split procedure in which Platt scaling is fit only on the validation set and evaluated on the test set. This would make the test ECE of 0.0715 a cleaner generalization statistic.
  2. [Section 5.3, Figures 13 and 14] The terminal-state assignments in the IBL/DL/NT clustering rely on the 'last projection before disruption,' defined as 0-5 ms before t_D (footnote 8). Since t_D is automatically computed by DEFUSE and no per-shot accuracy or uncertainty is reported, a type-dependent bias in t_D (e.g., detecting a precursor rather than the thermal quench onset) could create or distort the observed separation. Please add a sensitivity analysis that jitters t_D by at least ±10 ms and ideally ±50 ms and recomputes the state assignments, or report a per-disruption-type validation of the DEFUSE t_D against manual annotations.
  3. [Appendix A, Hyperparameter choice] The final model configuration was selected by manual inspection of latent spaces, and the authors note that 'we are likely selecting suboptimal settings.' To rule out that the central qualitative results (multimodal separation and the disruption-type clustering) are artifacts of a particular configuration, please report quantitative latent-space quality metrics for the chosen model and for a small neighborhood of the scanned hyperparameters. The mutual information used in Figure A.1, the reconstruction error, and a cluster-separability measure would be appropriate; showing that the main conclusions are stable across these configurations would considerably strengthen the paper.
minor comments (6)
  1. [Section 5.2, Figures 9 and 10] The notation for the disruptivity alternates between \hat{D}_disr in the text and D_disr in the figures and captions; please unify the notation.
  2. [Section 5.4] The threshold W1 ≥ 3 for selecting 'significant' feature differences is presented without supporting justification or multiple-testing correction; this is acceptable for a proof-of-principle, but the exploratory nature should be stated explicitly.
  3. [Section 3.1, Footnote 4] The footnote states that early evaluations indicate the aggregate distributions of event detections are statistically meaningful; please clarify how this was assessed, since these detections are used as evaluation metadata.
  4. [Figure 8] The claim of 'a notable lack of estimates between 1/16 to 1/2' would be more informative if accompanied by the fraction of samples in each bin of the reliability diagram.
  5. [Section 4.4, Table A.5] The sentence 'The output of the FNO layers is flattened and concatenated to the previous value of µ' is slightly misleading, since the architecture in Table A.5 uses a two-input MLP rather than a simple concatenation; please rephrase to match the implementation.
  6. [Section 4.3, Eq. (29)] Please define the domain of t_m explicitly and clarify that the ramp only applies for t_D - B ≤ t_m < t_D - A, with y_tm = 1 for t_m ≥ t_D - A, so that readers do not have to infer the piecewise structure.

Circularity Check

3 steps flagged · score 6.0 of 10

The reported Drisk calibration is a post hoc fit to the same tD-derived disruption rate used to build the training labels; DEFUSE supplies the unvalidated tD metadata.

  1. fitted input called prediction [Section 4.3 Eq. (29); Section 4.4 post-hoc calibration; Section 5.2 Eq. (37) and Figure 8]
    "For the disruption risk variable, we create labels ytm using the time of disruption tD as computed in [26]. [...] Then, we apply Platt scaling [67] to calibrate the prediction curve of Drisk to best represent the disruption rate. [...] We define this disruption rate as, for each location in z, the fraction of timeslices projected there where the discharge ends in a disruption."

    The risk head Drisk is trained with BCE against ytm, which is a deterministic function of tD (Eq. 29: 1 for tm >= tD-A, with a linear ramp from tD-B to tD-A). The evaluation target eDrate is defined as the fraction of timeslices at a latent location where the discharge ends in a disruption, which is also a function of tD. The paper then applies Platt scaling to Drisk 'to calibrate the prediction curve of Drisk to best represent the disruption rate'. The reported ECE therefore measures the error of a curve that was explicitly re-scaled to the same rate it is compared against.

  2. self citation load bearing [Section 3.1 dataset construction and Footnote 4]
    "We define tD as the onset of the thermal quench leading to the eventual disruption, and automatically compute these times using the DEFUSE framework [26]. [...] The development and evaluation of the automated event detection in DEFUSE is still ongoing. Nevertheless, early evaluations indicate that the aggregate distributions of detections are statistically meaningful."

    tD is the single most load-bearing input of the paper: it defines the risk labels (Eq. 29), the empirical disruption rate eDrate, the disruptivity map (last projections 0-5 ms before tD), and the terminal-state assignments for the IBL/DL/NT clustering. The paper's only support for tD is citation [26], the DEFUSE framework, which is authored by the same research group (Pau, Sauter, Poels, Venturini). Footnote 4 concedes that event-detection development is 'still ongoing' and offers only 'early evaluations' of aggregate distributions; no per-shot accuracy or uncertainty for tD is reported.

1 more flagged steps
  1. other [Section 5.1 hyperparameter selection (Figure A.1); Section 5.2 event correlations]
    "To quantify the benefit of adding more modes, we compute the mutual information [72] between the states and disruption-related event detections [26, 32]. [...] At 7 surrounding peaks, for a total of K = 8 Gaussians in the mixture prior, the benefit of adding more modes levels off [...] Additionally, we compute the correlations with disruption-related event detections from [26]."

    The number of prior modes K is selected by maximizing mutual information between the latent states and event detections from [26, 32]. The same event detections are then used in Section 5.2 to validate the states, so the reported event-state correlations are partly optimized by construction: K was chosen to make those correlations large. This is a selection-on-evaluation leak rather than a full circularity, and it does not affect the IBL/DL/NT separation, which uses external scenario labels.

full rationale

The VAE latent-space learning itself is self-contained: reconstruction, KL, uniformity, and risk losses are all defined in Eqs. (24)-(31), and the separation of IBL, density-limit, and negative-triangularity disruptions is validated against external scenario labels with no fitted parameters. However, the paper's headline quantitative claim, that Drisk 'accurately captures the real disruption rate' with ECE 0.0307/0.0715, is a calibrated goodness-of-fit rather than an independent prediction. The training labels ytm (Eq. 29) are a ramp to 1 computed from tD, and the 'real disruption rate' eDrate (Section 5.2) is the fraction of timeslices ending in disruption, also derived from tD; the post hoc Platt scaling is explicitly fit to eDrate. The train ECE is thus in-sample, and the test ECE only checks generalization of the calibrator to new shots, not the validity of the tD-based target. Additionally, tD and the event detections used in evaluation come from DEFUSE, a framework by the same authors whose event-detection development is admitted to be ongoing (Footnote 4), with no reported accuracy or uncertainty for tD. The selection of K by mutual information with the very event detections later used as validation adds a mild selection leak. The disruption-type separation (Figures 13-14), the component-plane physics correlations (Figure 15), and the counterfactual analysis (Section 5.4) are not circular and give the paper substantial independent content, which is why the overall score is 6 rather than higher.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a small number of modeling choices and data assumptions rather than on new physics. No invented physical entities are postulated. The main free parameters are the latent dimensionality, the number of prior modes, the disruption-label ramp, the loss weights, the prior geometry, and the post hoc calibration; several were selected by hand or by scanning against evaluation metadata. The key domain assumptions are the accuracy of DEFUSE times and events, the representativeness of the curated flat-top dataset, and the sufficiency of a 2D Markovian latent space.

free parameters (7)
  • Latent dimensionality d = 2
    Set to 2 for direct visualization; the authors state higher dimensions are future work. The 2D bottleneck is a deliberate modeling choice that shapes every latent-space result.
  • Number of prior modes K = 8
    Chosen by scanning K in [3,10] and selecting where mutual information between latent states and event detections levels off (Figure A.1). Uses evaluation metadata to choose model structure.
  • Disruption label ramp hyperparameters A and B = A = 0.15 s, B = 1.0 s
    Define ytm in Eq. 29; B is heuristic and A is tied to current redistribution time. They determine which portions of each discharge are labeled as near-disruption.
  • Loss weights a, b, c, d = a=2, b=5, c=100, d=10
    Tuned by hand (Appendix A). They balance reconstruction, KL, uniform mode coverage, and disruption-risk terms.
  • Prior and encoder distribution variances, radius, cluster scaling tau = sigma_p^2=0.1, sigma_phi^2=0.03, sigma_theta^2=0.05, radius=1.5, tau=0.05
    Chosen by manual scan; they control cluster separation, trajectory smoothness, and the softness of cluster assignment in Eq. A.2.
  • Time window w and stride s = w = 50, s = 50 at 10 kHz
    Fix the 5 ms latent trajectory resolution; no sensitivity analysis is reported.
  • Platt scaling parameters = fitted post hoc
    Fit to map raw Drisk to the empirical disruption rate; the reported ECE is computed after this fit.
assumptions (6)
  • domain assumption DEFUSE-derived disruption onset times tD and event detections are sufficiently accurate.
    These define the training labels in Eq. 29 and populate the evaluation metadata in Section 3.1; any systematic error propagates into both the risk head and the disruption-clustering evaluation.
  • domain assumption The curated TCV flat-top, lower single-null dataset is representative of the operational space being characterized.
    Section 3.1 filters to flat-top disruptions or regular terminations and notes a 70.4% disruption rate, so the absolute disruption rate is biased even if the relative structure is informative.
  • domain assumption A 2D Markovian latent trajectory with fixed 5 ms time windows can capture the global plasma state dynamics relevant to disruptions.
    Sections 2 and 4.2 make this modeling choice; the discussion concedes that fast, chain-of-events timescales are not represented.
  • domain assumption Post hoc Platt scaling is an appropriate calibration of Drisk to the empirical disruption rate.
    Section 5.2 assumes a monotone relationship between raw Drisk and the rate defined as the fraction of timeslices in a latent location that end in disruption.
  • domain assumption The automated confinement-state labels of reference [32] used at 75% or higher confidence are correct.
    Figure 11 uses these labels to interpret latent states; the cited classifier is unpublished and not formally verified in this paper.
  • standard math Standard VAE theory, including the ELBO and reparameterization, is valid as used.
    Sections 4.1 and 4.3 rely on references 20 and 21 without reproof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Plasma State Monitoring and Disruption Characterization using Multimodal VAEs." pith.science (2026). https://pith.science/paper/MWVK5YUO

@misc{pith2026250417710,
  author       = {Pith},
  title        = {Pith review of: Plasma State Monitoring and Disruption Characterization using Multimodal VAEs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MWVK5YUO}},
  note         = {Machine review of arXiv:2504.17710}
}
read the original abstract

When a plasma disrupts in a tokamak, significant heat and electromagnetic loads are deposited onto the surrounding device components. These forces scale with plasma current and magnetic field strength, making disruptions one of the key challenges for future devices. Unfortunately, disruptions are not fully understood, with many different underlying causes that are difficult to anticipate. Data-driven models have shown success in predicting them, but they only provide limited interpretability. On the other hand, large-scale statistical analyses have been a great asset to understanding disruptive patterns. In this paper, we leverage data-driven methods to find an interpretable representation of the plasma state for disruption characterization. Specifically, we use a latent variable model to represent diagnostic measurements as a low-dimensional, latent representation. We build upon the Variational Autoencoder (VAE) framework, and extend it for (1) continuous projections of plasma trajectories; (2) a multimodal structure to separate operating regimes; and (3) separation with respect to disruptive regimes. Subsequently, we can identify continuous indicators for the disruption rate and the disruptivity based on statistical properties of measurement data. The proposed method is demonstrated using a dataset of approximately 1600 TCV discharges, selecting for flat-top disruptions or regular terminations. We evaluate the method with respect to (1) the identified disruption risk and its correlation with other plasma properties; (2) the ability to distinguish different types of disruptions; and (3) downstream analyses. For the latter, we conduct a demonstrative study on identifying parameters connected to disruptions using counterfactual-like analysis. Overall, the method can adequately identify distinct operating regimes characterized by varying proximity to disruptions in an interpretable manner.

Figures

Figures reproduced from arXiv: 2504.17710 by the authors.

Figure 1
Figure 1. The distribution of the discharges’ dates, binned [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Distributions of key plasma parameters in the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Schematic overview of the model structure. The model consists of encoder distribution [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (15 more)
Figure 4
Figure 4. Figure 4: Depiction of the training and inference procedure of the proposed method. Data [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Precomputed deformation of a uniform space, used [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The probability density of the chosen prior [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: The learned disruption risk variable Drisk (left), overlaid on the posterior distribution (right). Regions with low and high disruption risk are spread throughout the latent space, with a zone of risk-free plasmas projected on the blue peak in the top right. describe t…
Figure 10
Figure 10. Figure 10: A comparison of Drisk and Dˆ disr. For low estimates of disruption risk (≤ 0.45), we find near-0 rates of disruptivity. That is, plasma regimes with low estimates for Drisk are distant from the actual onset of disruptions. As the risk increases, the disruptivity incre…
Figure 9
Figure 9. Figure 9: A projection of the disruptivity Dˆ disr (left), overlaid on the disruption risk Drisk (right). We define Dˆ disr as the number of disruptions per second for a given plasma property space. It is computed by summing the last projections of the disrupting discharges (0 t…
Figure 11
Figure 11. Figure 11: Correlation between high-confidence (≥ 0.75) confinement state detections computed using [32] and the states found in z. For each state we denote the fraction of time L, D or H-mode is detected for the total time spent in a state. Even though no confinement state labe…
Figure 13
Figure 13. Figure 13: Projections at the start of the flat top (circle) [PITH_FULL_IMAGE:figures/full_fig_p012_13.png]
Figure 12
Figure 12. Figure 12: Comparing the two most commonly detected [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 14
Figure 14. Figure 14: The assignment of the projections from Figure 13 [PITH_FULL_IMAGE:figures/full_fig_p012_14.png]
Figure 16
Figure 16. Figure 16: Automatically identified parameters correlating to [PITH_FULL_IMAGE:figures/full_fig_p013_16.png]
Figure 17
Figure 17. Figure 17: An example discharge from IBL scenario development experiments that ended in a disruption, TCV #64389, and 1 [PITH_FULL_IMAGE:figures/full_fig_p014_17.png]
Figure 19
Figure 19. Figure 19: Empirical distributions up to [PITH_FULL_IMAGE:figures/full_fig_p014_19.png]
Figure 20
Figure 20. Figure 20: Alteration of discharge projections just before [PITH_FULL_IMAGE:figures/full_fig_p015_20.png]
Figure 21
Figure 21. Figure 21: Projections of the altered trajectories illustrated [PITH_FULL_IMAGE:figures/full_fig_p015_21.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

100 extracted references · 44 canonical work pages

  1. [40]

    Towards Transparent and Accurate Plasma State Monitoring at JET

    Bürli, A., Pau, A., Koller, T., Sauter, O., and Contributors, J. Towards transparent and accurate plasma state monitoring at JET. arXiv preprint arXiv:2502.12182, 2025. arXiv:2502.12182, doi:10.48550/arXiv.2502.12182

    work page Pith review arXiv doi:10.48550/arxiv.2502.12182 2025

  2. [1]

    and Porosnicu, C

    Jepu, I., Matthews, G., Widdowson, A., Rubel, M., Fortuna-Zale´sna, E., Zdunek, J., Petersson, P., Thompson, V ., Dinca, P. and Porosnicu, C. et al. Beryllium melting and erosion on the upper dump plates in JET during three ITER-like wall cam- paigns. Nuclear Fusion, 59(8):086009, jun 2019. https:// dx.doi.org/10.1088/1741-4326/ab2076, doi:10. 1088/1741-4...

    work page doi:10.1088/1741-4326/ab2076 2019

  3. [2]

    and Fortuna- Zalesna, E

    Jepu, I., Widdowson, A., Matthews, G., Coad, J., Likonen, J., Brezinsek, S., Rubel, M., Pintsuk, G., Petersson, P. and Fortuna- Zalesna, E. et al. Overview of damage to beryllium limiters by unmitigated disruptions and runaway electrons in the JET tokamak with metal walls. Nuclear Fusion, 64(10):106047, sep

  4. [3]

    and Hartmann, N

    Lehnen, M., Aleynikova, K., Aleynikov, P., Campbell, D., Drewelow, P., Eidietis, N., Gasparyan, Y ., Granetz, R., Gribov, Y . and Hartmann, N. et al. Disruptions in ITER and strategies for their control and mitigation. Journal of Nuclear Materials, 463:39–48,

  5. [4]

    and Holcomb, C

    Strait, E., Barr, J., Baruzzo, M., Berkery, J., Buttery, R., de Vries, P., Eidietis, N., Granetz, R., Hanson, J. and Holcomb, C. et al. Progress in disruption prevention for ITER. Nuclear Fusion, 59(11):112012, jun 2019. https:// dx.doi.org/10.1088/1741-4326/ab15de, doi:10. 1088/1741-4326/ab15de

    work page doi:10.1088/1741-4326/ab15de 2019

  6. [5]

    Essay: Overcoming the obstacles to a magnetic fusion power plant

    Fasoli, A. Essay: Overcoming the obstacles to a magnetic fusion power plant. Phys. Rev. Lett., 130:220001, May 2023. https: //link.aps.org/doi/10.1103/PhysRevLett.130. 220001, doi:10.1103/PhysRevLett.130.220001

    doi:10.1103/physrevlett.130 2023

  7. [6]

    Tokamaks

    Wesson, J. Tokamaks. International Series of Monographs on Physics. Clarendon Press, Oxford, England, 3 edition, November 2003. https://books.google.ch/books? id=iPlAwZI6HIYC

    2003

  8. [7]

    and Eidietis, N

    Bandyopadhyay, I., Igochine, V ., Sauter, O., Sabbagh, S., Park, J.-K., Nardon, E., Villone, F., Maraschek, M., Pautasso, G. and Eidietis, N. et al. 3.3 Disruption Prediction. Chapter 4: MHD, Disruptions and Control Physics in Tokamaks. Submitted to Nucl. Fusion

Show all 100 references
  1. [8]

    and Gribov, Y

    Hender, T., Wesley, J., Bialek, J., Bondeson, A., Boozer, A., Buttery, R., Garofalo, A., Goodman, T., Granetz, R. and Gribov, Y . et al. Chapter 3: MHD stability, operational limits and disruptions. Nuclear Fusion , 47(6):S128, jun 2007. https: //dx.doi.org/10.1088/0029-5515/4...

    2007 doi

  2. [9]

    Statistical analysis of disruptions in JET

    de Vries, P., Johnson, M., Segui, I., and JET EFDA Con- tributors. Statistical analysis of disruptions in JET. Nuclear Fusion , 49(5):055011, apr 2009. https: //dx.doi.org/10.1088/0029-5515/49/5/055011, doi:10.1088/0029-5515/49/5/055011

    2009 doi

  3. [10]

    Survey of dis- ruption causes at JET

    de Vries, P., Johnson, M., Alper, B., Buratti, P., Hender, T., Koslowski, H., Riccardo, V ., and JET-EFDA Contributors. Survey of dis- ruption causes at JET. Nuclear Fusion, 51(5):053018, apr 2011. https://dx.doi.org/10.1088/0029-5515/51/5/ 053018, doi:10.1088/0029-5515/51/5/053018

    2011 doi

  4. [11]

    and Henriques, R

    Gerasimov, S., Abreu, P., Artaserse, G., Baruzzo, M., Buratti, P., Carvalho, I., Coffey, I., De La Luna, E., Hender, T. and Henriques, R. et al. Overview of disruptions with JET- ILW. Nuclear Fusion, 60(6):066028, may 2020. https:// dx.doi.org/10.1088/1741-4326/ab87b0, doi:10....

    2020 doi

  5. [12]

    Deep learning

    LeCun, Y ., Bengio, Y ., and Hinton, G. Deep learning. Nature, 521(7553):436–444, May 2015. doi:10.1038/ nature14539

    2015

  6. [13]

    A real-time machine learning-based disruption predictor in DIII-D

    Rea, C., Montes, K., Erickson, K., Granetz, R., and Tinguely, R. A real-time machine learning-based disruption predictor in DIII-D. Nuclear Fusion, 59(9):096016, jul 2019. https:// dx.doi.org/10.1088/1741-4326/ab28bf, doi:10. 1088/1741-4326/ab28bf

    2019 doi

  7. [14]

    and Ding, Y

    Zheng, W., Xue, F., Chen, Z., Chen, D., Guo, B., Shen, C., Ai, X., Wang, N., Zhang, M. and Ding, Y . et al. Disruption prediction for future tokamaks using parameter-based transfer learning. Communications Physics, 6(1):181, Jul 2023. doi: 10.1038/s42005-023-01296-9

    2023 doi

  8. [15]

    Hybrid deep-learning architecture for general disruption predic- tion across multiple tokamaks

    Zhu, J., Rea, C., Montes, K., Granetz, R., Sweeney, R., and Tinguely, R. Hybrid deep-learning architecture for general disruption predic- tion across multiple tokamaks. Nuclear Fusion, 61(2):026007, dec

  9. [16]

    Predicting disruptive instabilities in controlled fusion plasmas through deep learning

    Kates-Harbeck, J., Svyatkovskiy, A., and Tang, W. Predicting disruptive instabilities in controlled fusion plasmas through deep learning. Nature, 568(7753):526–531, Apr 2019. doi:10. 1038/s41586-019-1116-4

    2019

  10. [17]

    A., Gelfusa, M., 17 and JET Contributors

    Vega, J., Murari, A., Dormido-Canto, S., Rattá, G. A., Gelfusa, M., 17 and JET Contributors. Disruption prediction with artificial intelli- gence techniques in tokamak plasmas. Nature Physics, 18(7):741– 750, Jul 2022. doi:10.1038/s41567-022-01602-2

    2022 doi

  11. [18]

    Disruption prediction at JET through deep convolutional neu- ral networks using spatiotemporal information from plasma pro- files

    Aymerich, E., Sias, G., Pisano, F., Cannas, B., Carcangiu, S., Sozzi, C., Stuart, C., Carvalho, P., Fanni, A., and JET Contributors. Disruption prediction at JET through deep convolutional neu- ral networks using spatiotemporal information from plasma pro- files. Nuclear Fusio...

    2022 doi

  12. [19]

    Improvements in disruption prediction at ASDEX upgrade

    Aledda, R., Cannas, B., Fanni, A., Pau, A., Sias, G., and the ASDEX Upgrade Team. Improvements in disruption prediction at ASDEX upgrade. Fusion Engineering and Design, 96-97:698–702, 2015. Proceedings of the 28th Symposium On Fusion Technology (SOFT- 28). https://www.scienced...

    2015 doi

  13. [20]

    Kingma, D. P. and Welling, M. Auto-encoding variational bayes. In International Conference on Learning Represen- tations, 2014. https://openreview.net/forum?id= 33X9fd2-9FyZd

    2014

  14. [21]

    J., Mohamed, S., and Wierstra, D

    Rezende, D. J., Mohamed, S., and Wierstra, D. Stochastic backpropagation and approximate inference in deep generative models. In Proceedings of the 31st International Conference on Machine Learning , volume 32 of Proceedings of Machine Learning Research, pages 1278–1286, Bejin...

  15. [22]

    Dynamical variational autoencoders: A comprehensive review

    Girin, L., Leglaive, S., Bie, X., Diard, J., Hueber, T., and Alameda- Pineda, X. Dynamical variational autoencoders: A comprehensive review. Found. Trends Mach. Learn., 15(1–2):1–175, December

  16. [23]

    Dilokthanakul, N., Mediano, P. A. M., Garnelo, M., Lee, M. C. H., Salimbeni, H., Arulkumaran, K., and Shanahan, M. Deep unsupervised clustering with gaussian mixture variational autoencoders. arXiv preprint arXiv:1611.02648, 2017. arXiv: 1611.02648, doi:10.48550/arXiv.1611.02648

    2017 doi arXiv

  17. [24]

    Tomczak, J. M. Deep Generative Modeling . Springer In- ternational Publishing, 2022. http://dx.doi.org/ 10.1007/978-3-030-93158-2 , doi:10.1007/ 978-3-030-93158-2

    2022 doi

  18. [25]

    and Auriemma, F

    Duval, B., Abdolmaleki, A., Agostini, M., Ajay, C., Alberti, S., Alessi, E., Anastasiou, G., Andrèbe, Y ., Apruzzese, G. and Auriemma, F. et al. Experimental research on the TCV toka- mak. Nuclear Fusion, 64(11):112023, oct 2024. https:// dx.doi.org/10.1088/1741-4326/ad8361, d...

    2024 doi

  19. [26]

    and Joffrin, E

    Pau, A., Sauter, O., Sommariva, C., Poels, Y ., Venturini, C., Labit, B., Imbeaux, F., Litaudon, X., Falchetto, G. and Joffrin, E. et al. A modern framework to support disruption studies: the EUROfusion disruption database. In 29th Fusion Energy Conference (IAEA- FEC), 2023. h...

    2023

  20. [27]

    Vertical Instability Studies in the TCV Tokamak and Development and Application of Multimachine Real-Time Proximity Control Strategies

    Marchioni, S. Vertical Instability Studies in the TCV Tokamak and Development and Application of Multimachine Real-Time Proximity Control Strategies. PhD thesis, EPFL, Lausanne, 2024. https://infoscience.epfl.ch/handle/20.500. 14299/242254, doi:10.5075/epfl-thesis-10943

    2024 doi

  21. [28]

    P., Martin, Y

    Turri, G., Sauter, O., Porte, L., Alberti, S., Asp, E., Goodman, T. P., Martin, Y . R., Udintsev, V . S., and Zucca, C. The role of MHD in the sustainment of electron internal transport barriers and H-mode in TCV. Journal of Physics: Conference Series, 123(1):012038, jul 2008....

    2008 doi

  22. [29]

    and Février, O

    Labit, B., Sauter, O., Pütterich, T., Bagnato, F., Camenen, Y ., Coda, S., Contré, C., Coosemans, R., Eriksson, F. and Février, O. et al. Progress in the development of the ITER baseline scenario in TCV. Plasma Physics and Controlled Fusion, 66(2):025016, jan

  23. [30]

    and Lennholm, M

    Sieglin, B., Maraschek, M., Gude, A., Klossek, F., Felici, F., Bernert, M., Kudlacek, O., Pau, A., Piron, L. and Lennholm, M. et al. H-mode density limit disruption avoidance in ASDEX Upgrade, TCV and JET. Fusion Engineering and Design , 215:114961,

  24. [31]

    and Marinoni, A

    Coda, S., Merle, A., Sauter, O., Porte, L., Bagnato, F., Boedo, J., Bol- zonella, T., Février, O., Labit, B. and Marinoni, A. et al. Enhanced confinement in diverted negative-triangularity L-mode plasmas in TCV. Plasma Physics and Controlled Fusion, 64(1):014004, dec

  25. [32]

    Robust confinement state classification with uncertainty quantification through ensembled data-driven methods

    Poels, Y ., Venturini, C., Pau, A., Sauter, O., Menkovski, V ., the TCV team, and the WPTE team. Robust confinement state classification with uncertainty quantification through ensembled data-driven methods. arXiv preprint arXiv:2502.17397 , 2025. arXiv: 2502.17397, doi:10.485...

    2025 doi arXiv

  26. [33]

    M., Svensén, M., and Williams, C

    Bishop, C. M., Svensén, M., and Williams, C. K. I. GTM: The generative topographic mapping. Neural Computation, 10(1):215– 234, 1998. doi:10.1162/089976698300017953

    1998 doi

  27. [34]

    A machine learning approach based on generative topographic mapping for disruption preven- tion and avoidance at JET

    Pau, A., Fanni, A., Carcangiu, S., Cannas, B., Sias, G., Murari, A., Ri- mini, F., and the JET Contributors. A machine learning approach based on generative topographic mapping for disruption preven- tion and avoidance at JET. Nuclear Fusion, 59(10):106017, aug

  28. [36]

    Self-organized formation of topologically correct feature maps

    Kohonen, T. Self-organized formation of topologically correct feature maps. Biological Cybernetics, 43(1):59–69, Jan 1982. doi:10.1007/BF00337288

    1982 doi

  29. [37]

    Mapping of the ASDEX upgrade operational space for disruption prediction

    Aledda, R., Cannas, B., Fanni, A., Sias, G., and Pautasso, G. Mapping of the ASDEX upgrade operational space for disruption prediction. IEEE Transactions on Plasma Science , 40(3):570–576, 2012. doi:10.1109/TPS.2011.2174385

    2012

  30. [38]

    A self-organised partition of the high dimensional plasma parameter space for plasma disruption prediction

    Aymerich, E., Fanni, A., Pisano, F., Sias, G., Cannas, B., JET Contributors, and WPTE Team. A self-organised partition of the high dimensional plasma parameter space for plasma disruption prediction. Nuclear Fusion , 64(10):106063, sep

  31. [41]

    A., Berkery, J

    Sabbagh, S. A., Berkery, J. W., Park, Y . S., Butt, J., Riquezes, J. D., Bak, J. G., Bell, R. E., Delgado-Aparicio, L., Gerhardt, S. P. and Ham, C. J. et al. Disrup- tion event characterization and forecasting in tokamaks. Physics of Plasmas , 30(3):032506, 03 2023. https: //p...

    2023 doi

  32. [42]

    and Esposito, B

    Maraschek, M., Gude, A., Igochine, V ., Zohm, H., Alessi, E., Bernert, M., Cianfarani, C., Coda, S., Duval, B. and Esposito, B. et al. Path-oriented early reaction to approaching disruptions in ASDEX Upgrade and TCV in view of the future needs for ITER and DEMO. Plasma Physics...

  33. [43]

    Density limits in toroidal plasmas

    Greenwald, M. Density limits in toroidal plasmas. Plasma Physics and Controlled Fusion , 44(8):R27, jul 2002. https: //dx.doi.org/10.1088/0741-3335/44/8/201, doi: 10.1088/0741-3335/44/8/201

    2002 doi

  34. [44]

    A statistical approach for the automatic identification of the start of the chain of events leading to the disruptions at JET

    Aymerich, E., Fanni, A., Sias, G., Carcangiu, S., Cannas, B., Murari, A., Pau, A., and JET contributors. A statistical approach for the automatic identification of the start of the chain of events leading to the disruptions at JET. Nuclear Fusion, 61(3):036013, feb

  35. [46]

    Current Driven Ideal MHD Modes in a Toka- mak, chapter 4, pages 55–68

    Zohm, H. Current Driven Ideal MHD Modes in a Toka- mak, chapter 4, pages 55–68. John Wiley & Sons, Ltd,

  36. [47]

    Integrated real-time supervisory management for off-normal-event handling and feedback control of tokamak plasmas

    Vu, T., Felici, F., Galperti, C., Maraschek, M., Pau, A., Rispoli, N., Sauter, O., and Sieglin, B. Integrated real-time supervisory management for off-normal-event handling and feedback control of tokamak plasmas. IEEE Transactions on Nuclear Science, 68(8):1855–1861, 2021. do...

    2021

  37. [48]

    Tokamak equilibrium reconstruction code LIUQE and its real time implementation

    Moret, J.-M., Duval, B., Le, H., Coda, S., Felici, F., and Reimerdes, H. Tokamak equilibrium reconstruction code LIUQE and its real time implementation. Fusion Engineering and Design, 91:1–15,

  38. [50]

    A dimensionality reduction algorithm for mapping tokamak opera- tional regimes using a variational autoencoder (V AE) neural net- work

    Wei, Y ., Levesque, J., Hansen, C., Mauel, M., and Navratil, G. A dimensionality reduction algorithm for mapping tokamak opera- tional regimes using a variational autoencoder (V AE) neural net- work. Nuclear Fusion, 61(12):126063, nov 2021. https:// dx.doi.org/10.1088/1741-432...

    2021 doi

  39. [51]

    A recurrent variational autoencoder for speech enhancement

    Leglaive, S., Alameda-Pineda, X., Girin, L., and Horaud, R. A recurrent variational autoencoder for speech enhancement. In ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 371–375, 2020. doi:10.1109/ICASSP40776.2020.9053164

    2020

  40. [52]

    T., Nguyen, T

    Dang, H., Huu, T. T., Nguyen, T. M., and Ho, N. Beyond vanilla variational autoencoders: Detecting posterior collapse in condi- tional and hierarchical variational autoencoders. In The Twelfth International Conference on Learning Representations , 2024. https://openreview.net/...

    2024

  41. [53]

    Q., Rubanova, Y ., Bettencourt, J., and Duvenaud, D

    Chen, T. Q., Rubanova, Y ., Bettencourt, J., and Duvenaud, D. Neural ordinary differential equations. In Advances in Neural Information Processing Systems, volume 31, pages 6572–6583, 2018. https: //proceedings.neurips.cc/paper/2018/hash/ 69386f6bb1dfed68692a24c8686939b9-Abstr...

    2018

  42. [54]

    Minka, T. et al. Divergence measures and message passing. Technical report, Technical report, Microsoft Research, 2005. https:// miat.inrae.fr/AIGM/biblios/TR-2005-173.pdf

    2005

  43. [55]

    KL( q||p) is mode-seeking

    Jones, A. KL( q||p) is mode-seeking. Accessed 2025-03-

    2025

  44. [56]

    and Tuck, J

    Kruskal, M. and Tuck, J. L. The instability of a pinched fluid with a longitudinal magnetic field. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences, 245(1241):222–237, June 1958. http://dx.doi. org/10.1098/rspa.1958.0079, doi:10.1098/rsp...

    1958

  45. [57]

    Shafranov, V . D. On magnetohydrodynamical equilibrium configurations. Soviet Phys. JETP , V ol: 6, 03 1958. https: //www.osti.gov/biblio/4305963

    1958

  46. [58]

    B., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A

    Li, Z., Kovachki, N. B., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A. M., and Anandkumar, A. Fourier neural operator for parametric partial differential equations. In International Con- ference on Learning Representations, volume 9, 2021. https: //openreview.net/...

    2021

  47. [59]

    https://onlinelibrary.wiley.com/doi/ abs/10.1002/9783527677375.ch4, doi:https: //doi.org/10.1002/9783527677375.ch4

    doi

  48. [60]

    Fast dynamic 1D simulation of divertor plasmas with neural PDE surrogates

    Poels, Y ., Derks, G., Westerhof, E., Minartz, K., Wiesen, S., and Menkovski, V . Fast dynamic 1D simulation of divertor plasmas with neural PDE surrogates. Nuclear Fusion, 63(12):126012, sep

  49. [61]

    J., and Deisenroth, M

    Gopakumar, V ., Gray, A., Zanisi, L., Nunn, T., Pamela, S., Giles, D., Kusner, M. J., and Deisenroth, M. P. Calibrated physics-informed uncertainty quantification. arXiv preprint arXiv:2502.04406, 2025. arXiv:2502.04406, doi:10. 48550/arXiv.2502.04406

    2025 doi arXiv

  50. [62]

    https://www.sciencedirect.com/science/ article/pii/S0920379614005973, doi:https: //doi.org/10.1016/j.fusengdes.2014.09.019

    2014 doi

  51. [63]

    M., and Team, T

    Testa, D., Team, E. M., and Team, T. Manufacturing, installation, commissioning, and first results with the 3D low-temperature co- fired ceramic high-frequency magnetic sensors on the Tokamak à Configuration Variable. Review of Scientific Instruments , 91(8):081401, 08 2020. d...

    2020 doi

  52. [64]

    and Leibler, R

    Kullback, S. and Leibler, R. A. On information and sufficiency. The Annals of Mathematical Statistics , 22(1):79–86, 1951. http: //www.jstor.org/stable/2236703

    1951

  53. [65]

    J., Zhang, Y ., and Xiao, Y

    Xu, Z.-Q. J., Zhang, Y ., and Xiao, Y . Training behavior of deep neural network in frequency domain. In Gedeon, T., Wong, K. W., and Lee, M., editors, Neural Information Processing, pages 264–274, Cham, 2019. Springer International Publishing. http: //dx.doi.org/10.1007/978-3...

    2019 doi

  54. [66]

    P., Tancik, M., Barron, J

    Mildenhall, B., Srinivasan, P. P., Tancik, M., Barron, J. T., Ramamoorthi, R., and Ng, R. NeRF: representing scenes as neural radiance fields for view synthesis. Commun. ACM, 65(1):99–106, December 2021. doi:10.1145/3503250

    2021 doi

  55. [67]

    Platt, J. et al. Probabilistic outputs for support vector ma- chines and comparisons to regularized likelihood meth- ods. Advances in large margin classifiers , 10(3):61–74,

  56. [68]

    and Contreras, P

    Murtagh, F. and Contreras, P. Algorithms for hierarchical cluster- ing: an overview. WIREs Data Mining and Knowledge Discov- ery, 2(1):86–97, 2012. https://wires.onlinelibrary. wiley.com/doi/abs/10.1002/widm.53, doi:https: //doi.org/10.1002/widm.53

    2012 doi

  57. [69]

    and Halldórsson, M

    Chandra, B. and Halldórsson, M. M. Approximation algo- rithms for dispersion problems. Journal of Algorithms , 38(2):438–465, 2001. https://www.sciencedirect. com/science/article/pii/S0196677400911453, doi:https://doi.org/10.1006/jagm.2000.1145

    2001

  58. [70]

    https://andrewcharlesjones.github.io/ journal/klqp.html

  59. [71]

    and Sun, L

    Zheng, Z. and Sun, L. Disentangling latent space for V AE by label rel- evant/irrelevant dimensions. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), June

  60. [72]

    https://openaccess.thecvf.com/content_ CVPR_2019/html/Zheng_Disentangling_Latent_ Space_for_VAE_by_Label_RelevantIrrelevant_ Dimensions_CVPR_2019_paper.html

  61. [73]

    Werbos, P. J. Generalization of backpropagation with application to a recurrent gas market model. Neural Networks, 1(4):339–356,

  62. [74]

    Ob- taining well calibrated probabilities using bayesian binning

    Pakdaman Naeini, M., Cooper, G., and Hauskrecht, M. Ob- taining well calibrated probabilities using bayesian binning. Proceedings of the AAAI Conference on Artificial Intelli- gence, 29(1), Feb. 2015. https://ojs.aaai.org/index. php/AAAI/article/view/9602, doi:10.1609/aaai. v29i1.9602

    2015 doi

  63. [75]

    Causal explanation

    Lewis, D. Causal explanation. In Philosophical Papers, Volume II, pages 214–240. Oxford University Press, 1986. doi:10.1093/ 0195036468.001.0001

    1986

  64. [76]

    and Deisenroth, M

    Gopakumar, V ., Pamela, S., Zanisi, L., Li, Z., Gray, A., Brennand, D., Bhatia, N., Stathopoulos, G., Kusner, M. and Deisenroth, M. P. et al. Plasma surrogate modelling using fourier neural op- erators. Nuclear Fusion, 64(5):056025, apr 2024. https:// dx.doi.org/10.1088/1741-4...

    2024 doi

  65. [77]

    Kantorovich, L. V . Mathematical methods of organizing and planning production. Management Science, 6(4):366–422, July

  66. [78]

    and Kautz, J

    Vahdat, A. and Kautz, J. NV AE: A deep hierarchical vari- ational autoencoder. In Advances in Neural Information Processing Systems , volume 33, pages 19667–19679. Cur- ran Associates, Inc., 2020. https://proceedings. neurips.cc/paper_files/paper/2020/file/ e3b21256183cf7c2c7a...

    2020

  67. [79]

    and Pfister, J.-P

    Horvat, C. and Pfister, J.-P. Denoising normalizing flow. In Advances in Neural Information Process- ing Systems , volume 34, pages 9099–9111. Curran Associates, Inc., 2021. https://proceedings. neurips.cc/paper_files/paper/2021/file/ 4c07fe24771249c343e70c32289c1192-Paper.pdf

    2021

  68. [80]

    and Pentland, K

    Pamela, S., Carey, N., Brandstetter, J., Akers, R., Zanisi, L., Buchanan, J., Gopakumar, V ., Hoelzl, M., Huijsmans, G. and Pentland, K. et al. Neural-parareal: Self-improving acceleration of fusion mhd simulations using time-parallelisation and neural operators. Computer Phys...

  69. [81]

    https://www.sciencedirect.com/science/ article/pii/S001046552400314X, doi:https: //doi.org/10.1016/j.cpc.2024.109391

    2024

  70. [82]

    Cooley, J. W. and Tukey, J. W. An algorithm for the machine calculation of complex fourier series. Mathematics of Computation , 19(90):297–301, 1965. doi:10.1090/ s0025-5718-1965-0178586-1

    1965

  71. [83]

    Ivakhnenko, A. G. Polynomial theory of complex systems. IEEE Transactions on Systems, Man, and Cybernetics, SMC-1(4):364– 378, 1971. doi:10.1109/TSMC.1971.4308320

    1971

  72. [84]

    and Gimpel, K

    Hendrycks, D. and Gimpel, K. Gaussian error linear units (GELUs). arXiv preprint arXiv:1606.08415, 2016. arXiv:1606.08415, doi:10.48550/ARXIV.1606.08415. 20 Appendix A. Settings and hyperparameters Training. The model training procedure consists of minimizing the combined loss...

    2016 doi arXiv

  73. [90]

    and Antiga, L

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N. and Antiga, L. et al. PyTorch: An imperative style, high-performance deep learning library. In Advances in Neural Informa- tion Processing Systems , volume 32, 2019. htt...

    2019

  74. [91]

    Multivariate confidence calibration for object detec- tion

    Kuppers, F., Kronenberger, J., Shantia, A., and Haselhoff, A. Multivariate confidence calibration for object detec- tion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Work- shops, June 2020. https://openaccess.thecvf. com/content_...

    2020

  75. [92]

    Cover, T. M. and Thomas, J. A. Entropy, Relative Entropy, and Mutual Information , chapter 2, pages 13–55. John Wiley & Sons, Ltd, 2005. https://onlinelibrary.wiley.com/ doi/abs/10.1002/047174882X.ch2, doi:https:// doi.org/10.1002/047174882X.ch2

    2005 doi

  76. [93]

    DeGroot, M. H. and Fienberg, S. E. The comparison and evaluation of forecasters. The Statistician, 32(1/2):12, March

  77. [97]

    and Chiba, S

    Sakoe, H. and Chiba, S. Dynamic programming algorithm optimization for spoken word recognition. IEEE Transactions on Acoustics, Speech, and Signal Processing, 26(1):43–49, 1978. doi:10.1109/TASSP.1978.1163055

    1978

  78. [102]

    E., Kit, A., Poels, Y

    Järvinen, A. E., Kit, A., Poels, Y . R. J., Wiesen, S., Menkovski, V ., Frassinetti, L., Dunne, M., ASDEX Upgrade Team, and JET Contributors. Representation learning algorithms for inferring machine independent latent features in pedestals in JET and AUG. Physics of Plasmas, 3...

    2024 doi

  79. [103]

    E., Poels, Y

    Kit, A., Järvinen, A. E., Poels, Y . R. J., Wiesen, S., Menkovski, V ., Fischer, R., Dunne, M., and ASDEX Upgrade Team. On learning latent dynamics of the AUG plasma state. Physics of Plasmas, 31(3):032504, 03 2024. doi:10.1063/5.0174128

    2024 doi

  80. [104]

    and Perek, A

    Galperti, C., Felici, F., Vu, T., Sauter, O., Carpanese, F., Kong, M., Marceca, G., Merle, A., Pau, A. and Perek, A. et al. Overview of the TCV digital real-time plasma control system and its applications. Fusion Engineering and Design , 208:114640,

  81. [105]

    https://www.sciencedirect.com/science/ article/pii/S0920379624004915, doi:https: //doi.org/10.1016/j.fusengdes.2024.114640

    2024

  82. [106]

    A., Khaled, A., Mishchenko, K., Mehta, H., and Cutkosky, A

    Defazio, A., Yang, X. A., Khaled, A., Mishchenko, K., Mehta, H., and Cutkosky, A. The road less scheduled. In The Thirty- eighth Annual Conference on Neural Information Processing Systems, 2024. https://openreview.net/forum?id= 0XeNkkENuI

    2024

  83. [1960]

    http://dx.doi.org/10.1287/mnsc.6.4.366, doi:10.1287/mnsc.6.4.366

    doi

  84. [1983]

    http://dx.doi.org/10.2307/2987588, doi: 10.2307/2987588

    doi

  85. [1988]

    https://www.sciencedirect.com/science/ article/pii/089360808890007X, doi:https: //doi.org/10.1016/0893-6080(88)90007-X

  86. [1999]

    https://www.bibsonomy.org/bibtex/ 1c5df9f9137085cad9cafce3c347b2508

  87. [2014]

    https://proceedings.mlr.press/v32/ rezende14.html

    PMLR. https://proceedings.mlr.press/v32/ rezende14.html

  88. [2015]

    https://www.sciencedirect.com/science/ article/pii/S0022311514007594, doi:https: //doi.org/10.1016/j.jnucmat.2014.10.075

    2014 doi

  89. [2017]

    https://dx.doi.org/10.1088/1361-6587/ aa8d05, doi:10.1088/1361-6587/aa8d05

    doi

  90. [2021]

    doi:10.1561/2200000089

    doi

  91. [2023]

    https://dx.doi.org/10.1088/1741-4326/ acf70d, doi:10.1088/1741-4326/acf70d

    doi

  92. [2025]

    https://www.sciencedirect.com/science/ article/pii/S0920379625001619, doi:https: //doi.org/10.1016/j.fusengdes.2025.114961

    2025

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.