arxiv: 2605.14939 · v1 · submitted 2026-05-14 · ⚛️ physics.plasm-ph · cs.LG

Recognition: no theorem link

Real-time virtual circuits for plasma shape control via neural network emulators

Alasdair Ross , George K. Holt , Kamran Pentland , Adriano Agnello , Nicola C. Amorisco , Pedro Cavestany , Aran Garrod , Timothy Nunn

show 2 more authors

Charles Vincent Graham McArdle

Authors on Pith no claims yet

Pith reviewed 2026-05-15 14:24 UTC · model grok-4.3

classification ⚛️ physics.plasm-ph cs.LG

keywords tokamak plasma controlvirtual circuitsneural network emulatorsGrad-Shafranov equilibriareal-time shape controlMAST Upgradeplasma equilibria simulation

0 comments

The pith

Neural network emulators produce accurate real-time virtual circuits to control tokamak plasma shape.

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

The paper develops neural networks trained on simulated plasma equilibria to emulate shape parameters and derive virtual circuits on the fly. These circuits allow independent control of coupled shape parameters without relying on precomputed schedules that only work near reference points. By providing differentiable functions, the emulators enable rapid calculation of control vectors that maintain orthogonality across many different plasma states. This approach addresses the limitation of traditional methods in handling rapidly changing plasma configurations during experiments.

Core claim

By training neural networks on over a million simulated Grad-Shafranov equilibria, the authors create emulators from which virtual circuits can be derived in real time. These emulated circuits maintain high accuracy and orthogonality across diverse plasma states, validating their use as a general replacement for fixed schedules of precomputed circuits in the MAST-U control system.

What carries the argument

Neural network emulators of shape parameters, whose gradients yield the virtual circuits that decouple control actions for a given equilibrium.

Load-bearing premise

The neural networks, trained solely on computer-generated plasma equilibria, will generalize to produce usable virtual circuits on real MAST-U plasmas.

What would settle it

Direct comparison of shape control performance in MAST-U experiments using emulated versus precomputed virtual circuits, particularly for plasmas far from reference states, would show if accuracy holds.

Figures

Figures reproduced from arXiv: 2605.14939 by Adriano Agnello, Alasdair Ross, Aran Garrod, Charles Vincent, George K. Holt, Graham McArdle, Kamran Pentland, Nicola C. Amorisco, Pedro Cavestany, Timothy Nunn.

**Figure 2.** Figure 2: Two-dimensional UMAP embedding derived from all seven output dimensions for a random [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Kernel density estimation (KDE) plots for a selection of shape parameters for a sample of [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Histograms of realised shifts, as in eq. ( [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

**Figure 5.** Figure 5: Distributions of the input features used to train the NN emulators. [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗

**Figure 6.** Figure 6: Distributions of the output shape parameters used to train the NN emulators. These [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗

**Figure 7.** Figure 7: Two-dimensional coverage of the training dataset for every model input and output pair. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗

**Figure 8.** Figure 8: Pairwise Pearson correlation coefficients among the output shape parameters in the train [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗

**Figure 9.** Figure 9: Ensemble model performance evaluated on the test set. On the left are histograms of [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗

**Figure 10.** Figure 10: Histograms of realised shifts in eq. (12), normalised to the requested shift δPreq. This plot compares finite-difference GS derivatives (blue) with emulator VCs from the ensemble (orange) and the top model (red) evaluated on early-chain equilibria (MCMC step < 50) within the test set. This shows that the GS VCs perform best, and that an ensemble of models performs better than the single top model. 22 [PI… view at source ↗

**Figure 11.** Figure 11: Histograms of realised shifts in eq. (12), normalised to the requested shift δPreq. This plot compares emulator VCs computed via finite-difference derivatives with different step sizes (blue, green and red) with automatic differentiation (orange) evaluated on early-chain equilibria (MCMC step < 50) within the test set. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗

read the original abstract

Reliable position and shape control in tokamak plasmas requires accurate real-time regulation of several strongly coupled shape parameters. The control vectors that disentangle these couplings, referred to as \textit{virtual circuits} (VCs), enable independent shape parameter control for a specific Grad--Shafranov (GS) equilibrium. Numerical calculation of VCs is not currently feasible in real time, therefore VCs are usually computed prior to each experiment, using a small number of reference GS equilibria sampled along the desired scenario trajectory, with each VC used to control the plasma within a preset time interval. While effective near the reference equilibrium, this approach can lead to degraded performance as the plasma departs from the reference equilibrium and/or from the desired trajectory, and it complicates the design of robust control strategies for rapidly evolving plasma configurations. In this paper, we construct neural-network-based emulators of plasma shape parameters from which VCs can be derived, to provide the MAST Upgrade (MAST-U) plasma control system with state-aware VCs in real-time. To do this, we develop an extensive library of over a million simulated GS equilibria, covering a substantial portion of the MAST-U operational space. These emulators provide differentiable functions whose gradients can be rapidly computed, enabling the derivation of accurate VCs for real-time shape control. We perform extensive verification of the emulated VCs by testing whether they disentangle the control problem. The neural-network-based approach delivers high accuracy and orthogonality across a diverse range of equilibria. This work establishes the physical validity of emulated VCs as a scalable and general alternative to schedules of precomputed VCs.

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

Neural emulators from a large simulated GS library can generate real-time state-aware virtual circuits for MAST-U, but the work stays inside simulation and skips real-plasma checks.

read the letter

The main takeaway is that neural emulators built from a million-plus library of simulated equilibria let you compute state-dependent virtual circuits fast enough for real-time use on MAST-U. This sidesteps the current practice of locking in a few reference VCs for fixed time windows and should allow better handling of evolving plasma shapes during a discharge. They do a good job laying out why the precomputed approach falls short when the plasma drifts away from the chosen references and showing how the differentiable emulators fix that by providing gradients on demand. Building such a large library to cover a substantial portion of the operational space is a sensible move for robustness. The verification that the resulting circuits keep the shape parameters orthogonal across many test equilibria is a solid first check on whether the method works as intended, and it directly tests the disentangling property that matters for control. Where it is thinner is the jump from simulation to experiment. The tests use equilibria drawn from the training distribution, so they do not address how well the networks handle the differences that appear in real MAST-U shots, such as non-ideal effects not captured in the pure Grad-Shafranov model. The abstract also gives no concrete error metrics or tolerance numbers, which leaves the high accuracy statement without a clear benchmark for judging control performance. This paper is aimed at fusion plasma control engineers who need more adaptable real-time tools. It deserves a serious referee because the underlying idea is practical and the scale of the data generation is substantial; reviewers can ask for the missing experimental comparisons and quantitative results in the next round. The work shows honest engagement with the control problem even if the final validation step remains ahead.

Referee Report

2 major / 2 minor

Summary. The paper develops neural-network emulators trained on a library of over one million simulated Grad-Shafranov equilibria to compute differentiable plasma shape parameters for MAST-U. From these emulators, virtual circuits (VCs) are derived via gradients to enable real-time, state-aware shape control that disentangles coupled parameters, replacing schedules of precomputed VCs based on a few reference equilibria. Verification tests on simulated equilibria are reported to demonstrate high accuracy and orthogonality across a range of configurations.

Significance. If the emulators prove robust, the work would offer a scalable route to adaptive real-time control for evolving tokamak plasmas, reducing reliance on limited reference equilibria and enabling more robust strategies for dynamic scenarios. The scale of the simulation library and the use of differentiable emulators for gradient-based VC derivation are notable strengths that support reproducibility and extensibility.

major comments (2)

[Abstract and verification section] Abstract and verification section: the central claim of 'high accuracy and orthogonality' and 'physical validity' for MAST-U rests on verification performed exclusively on simulated equilibria drawn from the training distribution. No quantitative metrics (e.g., RMS errors, orthogonality measures, or error distributions), test protocol details, or comparisons against experimental equilibrium reconstructions from MAST-U discharges are provided, leaving the transfer assumption untested.
[Verification tests] Verification tests: the reported disentanglement succeeds only for ideal GS solutions; real plasmas include kinetic, resistive, and wall-eddy effects absent from the training data. This gap directly affects the load-bearing claim that emulated VCs constitute a 'scalable and general alternative' for experimental use, as no evidence of robustness under these perturbations is shown.

minor comments (2)

[Abstract] The abstract would benefit from inclusion of at least one quantitative performance figure (e.g., mean shape-parameter error or orthogonality metric) to substantiate the 'high accuracy' assertion.
[Methods] Clarify the precise definition and mathematical construction of virtual circuits (including how gradients are extracted from the emulator) in the methods section to aid readers outside the immediate subfield.

Simulated Author's Rebuttal

2 responses · 2 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major point below, clarifying the verification scope and committing to revisions that strengthen the presentation without overstating the current results.

read point-by-point responses

Referee: [Abstract and verification section] Abstract and verification section: the central claim of 'high accuracy and orthogonality' and 'physical validity' for MAST-U rests on verification performed exclusively on simulated equilibria drawn from the training distribution. No quantitative metrics (e.g., RMS errors, orthogonality measures, or error distributions), test protocol details, or comparisons against experimental equilibrium reconstructions from MAST-U discharges are provided, leaving the transfer assumption untested.

Authors: We agree that the abstract should explicitly report quantitative metrics. In the revised version we will insert specific RMS error values, orthogonality measures, and summary error statistics drawn from the verification tests already performed on the simulated equilibria. The full verification section contains the test protocol details and error distributions; we will add cross-references to make these more prominent. Direct comparisons against experimental MAST-U reconstructions are not present because the work is scoped to ideal Grad-Shafranov emulators; such comparisons constitute a separate validation step that we flag for future study. revision: partial
Referee: [Verification tests] Verification tests: the reported disentanglement succeeds only for ideal GS solutions; real plasmas include kinetic, resistive, and wall-eddy effects absent from the training data. This gap directly affects the load-bearing claim that emulated VCs constitute a 'scalable and general alternative' for experimental use, as no evidence of robustness under these perturbations is shown.

Authors: The training library and all verification results are restricted to ideal Grad-Shafranov equilibria, which is the standard basis for real-time magnetic shape control. The emulated VCs therefore provide state-aware disentanglement within that ideal framework. We will add a new limitations subsection that explicitly discusses the absence of kinetic, resistive, and wall-eddy effects and outlines how the emulator architecture could be extended by retraining on augmented libraries that incorporate these physics. No robustness data under those perturbations can be supplied from the existing ideal-GS dataset. revision: partial

standing simulated objections not resolved

Direct quantitative comparison of emulated VCs against experimental MAST-U equilibrium reconstructions
Robustness verification of the emulated VCs under kinetic, resistive, and wall-eddy perturbations

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper builds a library of over one million simulated Grad-Shafranov equilibria, trains neural-network emulators on them, and derives virtual circuits from the gradients of the resulting differentiable functions. All verification of accuracy and orthogonality is performed on simulated equilibria drawn from the same distribution. No step reduces a claimed prediction or result to a fitted input by construction, no load-bearing premise rests on self-citation chains, and no uniqueness theorem or ansatz is imported from prior author work. The workflow is therefore self-contained as a synthetic-data modeling and testing pipeline; the transfer assumption to real MAST-U plasmas is stated as an external claim rather than a derived quantity.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that simulated Grad-Shafranov equilibria sufficiently represent real MAST-U plasmas and that neural networks trained on them will produce accurate, differentiable emulators usable for control.

free parameters (1)

Neural network architecture and training hyperparameters
Choices of network depth, width, activation functions, and optimization settings are selected to fit the simulated equilibrium library.

axioms (1)

domain assumption The Grad-Shafranov equation provides an accurate model of axisymmetric tokamak equilibria
All training data are generated from numerical solutions of this equation.

pith-pipeline@v0.9.0 · 5633 in / 1323 out tokens · 54046 ms · 2026-05-15T14:24:13.564776+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 1 internal anchor

[1]

Ariola and A

M. Ariola and A. Pironti.Magnetic Control of Tokamak Plasmas. Advances in Industrial Control. Springer London, 2008

work page 2008
[2]

Walker, Peter De Vries, Federico Felici, and Eugenio Schuster

Michael L. Walker, Peter De Vries, Federico Felici, and Eugenio Schuster. Introduction to tokamak plasma control. In2020 American Control Conference (ACC), pages 2901–2918, 2020

work page 2020
[3]

Design and implementation of a model-based hierarchical architecture for plasma shape control in the tcv tokamak

A Mele, A Tenaglia, F Felici, C Galperti, D Carnevale, S Coda, A Merle, A Pironti, O Sauter, the TCV team, and the Eurofusion Tokamak Exploitation team. Design and implementation of a model-based hierarchical architecture for plasma shape control in the tcv tokamak. Plasma Physics and Controlled Fusion, 67(6):065035, jun 2025

work page 2025
[4]

The mast upgrade plasma control system.Fusion Engineering and Design, 159:111764, 2020

Graham McArdle, Luigi Pangione, and Martin Kochan. The mast upgrade plasma control system.Fusion Engineering and Design, 159:111764, 2020

work page 2020
[5]

J. T. Wai, M. D. Boyer, D. J. Battaglia, F. Carpanese, F. Felici, W. P. Wehner, A. S. Welander, and E. Kolemen. A tutorial on inversion-based shape control with design application to nstx-u, 2026

work page 2026
[6]

Pentland, N

K. Pentland, N. C. Amorisco, A. Ross, P. Cavestany, T. Nunn, A. Agnello, G. K. Holt, G. McArdle, C. Vincent, J. Buchanan, and S. J. P. Pamela. The FreeGSNKE Pulse Design Tool (FPDT): a computational framework for evolutive plasma scenario and control design, 2026

work page 2026
[7]

A neural network approach to tokamak equilibrium control

Chris Bishop, Peter Cox, Paul Haynes, Colin Roach, Mike Smith, Tom Todd, and David Trotman. A neural network approach to tokamak equilibrium control. In J. G. Taylor, editor, Neural Network Applications, pages 114–128, London, 1992. Springer London

work page 1992
[8]

Bishop, Paul S

Chris M. Bishop, Paul S. Haynes, Mike E. U. Smith, Tom N. Todd, and David L. Trotman. Real-time control of a tokamak plasma using neural networks.Neural Computation, 7(1):206–217, 1995

work page 1995
[9]

J. T. Wai, M. D. Boyer, and E. Kolemen. Neural net modeling of equilibria in NSTX-U. Nuclear Fusion, 62(8):086042, 2022

work page 2022
[10]

L. L. Lao, S. Kruger, C. Akcay, P. Balaprakash, T. A. Bechtel, E. Howell, J. Koo, J. Leddy, M. Leinhauser, Y. Q. Liu, S. Madireddy, J. McClenaghan, D. Orozco, A. Pankin, D. Schissel, S. Smith, X. Sun, and S. Williams. Application of machine learning and artificial intelligence to extend EFIT equilibrium reconstruction.Plasma Physics and Controlled Fusion,...

work page 2022
[11]

Prokhorov, Yuri V

Artem A. Prokhorov, Yuri V. Mitrishkin, Pavel S. Korenev, and Mikhail I. Patrov. The plasma shape control system in the tokamak with the artificial neural network as a plasma equilibrium reconstruction algorithm.IFAC-PapersOnLine, 53(2):857–862, 2020. 21st IFAC World Congress

work page 2020
[12]

Rui et al

W. Rui et al. Adaptive vertical position control system based on neural networks.Nuclear Fusion, 66(2):026012, 2025

work page 2025
[13]

Rasouli, C

H. Rasouli, C. Rasouli, and A. Koohi. Identification and control of plasma vertical position using neural network in Damavand tokamak.Review of Scientific Instruments, 84(2):023504, 2013

work page 2013
[14]

Adaptive vertical position control system based on neural networks.Nuclear Fusion, 66(2):026012, February 2026

Wangyi Rui, Yuehang Wang, Huihui Song, Zhongmin Huang, Zhengping Luo, Yao Huang, Zijie Liu, Kai Wu, Junjie Huang, and Bingjia Xiao. Adaptive vertical position control system based on neural networks.Nuclear Fusion, 66(2):026012, February 2026

work page 2026
[15]

Neural network-based confinement mode prediction for real-time disruption avoidance.IEEE Transactions on Plasma Science, 50(11):4157–4164, 2022

David Orozco, Brian Sammuli, Jayson Barr, William Wehner, and David Humphreys. Neural network-based confinement mode prediction for real-time disruption avoidance.IEEE Transactions on Plasma Science, 50(11):4157–4164, 2022

work page 2022
[16]

Implementation of ai/deep learning disruption predictor into a plasma control system.Contributions to Plasma Physics, 63(5-6):e202200095, 2023

William Tang, Ge Dong, Jayson Barr, Keith Erickson, Rory Conlin, Dan Boyer, Julian Kates-Harbeck, Kyle Felker, Cristina Rea, Nikolas Logan, Alexey Svyatkovskiy, Eliot Feibush, Joseph Abbatte, Mitchell Clement, Brian Grierson, Raffi Nazikian, Zhihong Lin, David Eldon, Auna Moser, and Mikhail Maslov. Implementation of ai/deep learning disruption predictor i...

work page 2023
[17]

Sutton and Andrew G

Richard S. Sutton and Andrew G. Barto.Reinforcement learning - an introduction. Adaptive computation and machine learning. MIT Press, 1998

work page 1998
[18]

Gianmaria De Tommasi, Sara Dubbioso, Yao Huang, Zheng-Ping Luo, Adriano Mele, and B. J. Xiao. A RL-based vertical stabilization system for the EAST tokamak. In2022 American Control Conference (ACC), pages 5328–5333, Atlanta, GA, USA, 2022

work page 2022
[19]

Magnetic control of tokamak plasmas through deep reinforcement learning.Nature, 602:414–419, 2022

Jonas Degrave, Federico Felici, Jonas Buchli, Michael Neunert, Brendan Tracey, Francesco Carpanese, Timo Ewalds, Roland Hafner, Abbas Abdolmaleki, Diego de las Casas, Craig Donner, Leslie Fritz, Cristian Galperti, Andrea Huber, James Keeling, Maria Tsimpoukelli, Jackie Kay, Antoine Merle, Jean-Marc Moret, Seb Noury, Federico Pesamosca, David Pfau, Olivier...

work page 2022
[20]

Avoiding fusion plasma tearing instability with deep reinforcement learning.Nature, 626(8000):746–751, 2024

Jaemin Seo, SangKyeun Kim, Azarakhsh Jalalvand, Rory Conlin, Andrew Rothstein, Joseph Abbate, Keith Erickson, Josiah Wai, Ricardo Shousha, and Egemen Kolemen. Avoiding fusion plasma tearing instability with deep reinforcement learning.Nature, 626(8000):746–751, 2024

work page 2024
[21]

Brendan D. Tracey, Andrea Michi, Yuri Chervonyi, Ian Davies, Cosmin Paduraru, Nevena Lazic, Federico Felici, Timo Ewalds, Craig Donner, Cristian Galperti, Jonas Buchli, Michael Neunert, Andrea Huber, Jonathan Evens, Paula Kurylowicz, Daniel J. Mankowitz, Martin Riedmiller, and The TCV Team. Towards practical reinforcement learning for tokamak magnetic con...

work page 2024
[22]

Nardon, and Philippe Moreau

Samy Kerboua-Benlarbi, R´ emy Nouailletas, Blaise Faugeras, E. Nardon, and Philippe Moreau. Magnetic control of WEST plasmas through deep reinforcement learning.IEEE Transactions on Plasma Science, 52:3698–3703, 2024

work page 2024
[23]

Offline model-based reinforcement learning for tokamak control

Ian Char, Joseph Abbate, Laszlo Bardoczi, Mark Boyer, Youngseog Chung, Rory Conlin, Keith Erickson, Viraj Mehta, Nathan Richner, Egemen Kolemen, and Jeff Schneider. Offline model-based reinforcement learning for tokamak control. In Nikolai Matni, Manfred Morari, and George J. Pappas, editors,Proceedings of The 5th Annual Learning for Dynamics and Control ...

work page 2023
[24]

L.L. Lao, H. St. John, R.D. Stambaugh, A.G. Kellman, and W. Pfeiffer. Reconstruction of current profile parameters and plasma shapes in tokamaks.Nuclear Fusion, 25(11):1611, nov 1985

work page 1985
[25]

Pentland, N

K. Pentland, N. C. Amorisco, O. El-Zobaidi, S. Etches, A. Agnello, G. K. Holt, A. Ross, C. Vincent, J. Buchanan, S. Pamela, G. McArdle, L. Kogan, and G. Cunningham. Validation of the static forward Grad–Shafranov equilibrium solvers in FreeGSNKE and Fiesta using EFIT++ reconstructions from MAST-U.Physica Scripta, 2024

work page 2024
[26]

Kochan, H

M. Kochan, H. Anand, A. Lvovskiy, P. Ryan, K. Verhaegh, T. Wijkamp, A. Kirk, and G. McArdle. Real-time plasma shape reconstruction on MAST Upgrade based on local expansion. In30th IEEE Symposium on Fusion Engineering (SOFE), Oxford, UK, 2023. Conference presentation, 9–13 July 2023

work page 2023
[27]

Anand, W

H. Anand, W. Wehner, D. Eldon, A. Welander, Z. Xing, A. Lvovskiy, J. Barr, E. Cho, B. Sammuli, D. Humphreys, N. Eidietis, A. Leonard, M. Kochan, C. Vincent, G. McArdle, G. Cunningham, A. Thornton, J. Harrison, V. Soukhanovskii, and J. Lovell. Real-time plasma equilibrium reconstruction and shape control for the mast upgrade tokamak.Nuclear Fusion, 64(8):0...

work page 2024
[28]

Amorisco, George K

Pedro Cavestany, Alasdair Ross, Adriano Agnello, Aran Garrod, Nicola C. Amorisco, George K. Holt, Kamran Pentland, and James Buchanan. Real-time applicability of emulated virtual circuits for tokamak plasma shape control. In2025 IEEE Conference on Control Technology and Applications (CCTA), pages 826–831, 2025

work page 2025
[29]

G. Cybenko. Approximation by superpositions of a sigmoidal function.Mathematics of Control, Signals and Systems, 2(4):303–314, 1989. 29 Journalvv(yyyy) aaaaaa Rosset al

work page 1989
[30]

Multilayer feedforward networks are universal approximators.Neural Networks, 2(5):359–366, 1989

Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators.Neural Networks, 2(5):359–366, 1989

work page 1989
[31]

Lin, Allan Pinkus, and Shimon Schocken

Moshe Leshno, Vladimir Ya. Lin, Allan Pinkus, and Shimon Schocken. Multilayer feedforward networks with a nonpolynomial activation function can approximate any function.Neural Networks, 6(6):861–867, 1993

work page 1993
[32]

Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks.Neural Networks, 3(5):551–560, 1990

Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks.Neural Networks, 3(5):551–560, 1990

work page 1990
[33]

Approximation capabilities of multilayer feedforward networks.Neural Networks, 4(2):251–257, 1991

Kurt Hornik. Approximation capabilities of multilayer feedforward networks.Neural Networks, 4(2):251–257, 1991

work page 1991
[34]

N. C. Amorisco, A. Agnello, G. Holt, M. Mars, J. Buchanan, and S. Pamela. Freegsnke: A python-based dynamic free-boundary toroidal plasma equilibrium solver.Physics of Plasmas, 31(4):042517, 04 2024

work page 2024
[35]

Agnello, N

A. Agnello, N. C. Amorisco, A. Keats, G. K. Holt, J. Buchanan, S. Pamela, C. Vincent, and G. McArdle. Emulation techniques for scenario and classical control design of tokamak plasmas.Physics of Plasmas, 31(4):043901, 04 2024

work page 2024
[36]

Umap: Uniform manifold approximation and projection.The Journal of Open Source Software, 3(29):861, 2018

Leland McInnes, John Healy, Nathaniel Saul, and Lukas Grossberger. Umap: Uniform manifold approximation and projection.The Journal of Open Source Software, 3(29):861, 2018

work page 2018
[37]

Optuna: A next-generation hyperparameter optimization framework

Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. InThe 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2623–2631, 2019

work page 2019
[38]

Pentland, A

K. Pentland, A. Ross, N. C. Amorisco, P. Cavestany, T. Nunn, A. Agnello, G. K. Holt, and C. Vincent. Real-time virtual circuits for plasma shape control via neural network surrogates: dynamic validation in closed-loop simulations, 2026

work page 2026
[39]

Sobolev Training for Neural Networks

Wojciech Marian Czarnecki, Simon Osindero, Max Jaderberg, Grzegorz Swirszcz, and Razvan Pascanu. Sobolev training for neural networks.CoRR, abs/1706.04859, 2017. 30

work page internal anchor Pith review Pith/arXiv arXiv 2017