pith. sign in

arxiv: 2606.27907 · v1 · pith:FOEZEPO3new · submitted 2026-06-26 · 🪐 quant-ph

End-to-End Learning of Quantum Control on Latent Dynamical Manifold

Jun-Dong Zhong , Zong-Yuan Ge , Feng-Hua Ren , Zhao-Ming Wang This is my paper

Pith reviewed 2026-06-29 04:27 UTC · model grok-4.3

classification 🪐 quant-ph
keywords quantum controllatent manifoldLSTMend-to-end learningopen quantum systemsadiabatic speedupstate transfernoise resilience
0
0 comments X

The pith

An LSTM on a learned latent manifold maps initial quantum states and noise parameters directly to trajectories and control pulses in one forward pass.

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

The paper establishes an end-to-end framework that replaces the traditional decoupled simulation-plus-optimization loop with a single neural network pass. System dynamics and optimal controls are learned jointly inside a low-dimensional latent manifold produced by an LSTM. On two concrete tasks—adiabatic speedup of a two-level system and state transfer in a noisy spin chain—the model produces both accurate trajectories and high-fidelity pulses while cutting optimization cost by three orders of magnitude. The same network generalizes across varying noise realizations, initial states, and driving fields. A reader should care because the method removes the main computational bottleneck that currently prevents real-time adaptive control of open quantum systems.

Core claim

The central claim is that a long short-term memory network trained on trajectories can embed the relevant open-system dynamics into a low-dimensional latent manifold, from which both future states and the corresponding optimized control pulses can be decoded directly from initial conditions and environmental parameters, eliminating iterative optimization.

What carries the argument

LSTM encoder-decoder operating on a low-dimensional latent dynamical manifold that jointly represents open-system evolution and control strategy.

If this is right

  • Fidelity increases on both the adiabatic speedup task and the spin-chain state transfer task relative to conventional methods.
  • Optimization cost drops by three orders of magnitude because no iterative simulation loop is required after training.
  • The network generalizes to multi-parameter time-varying noise, unseen initial states, and different driving fields without retraining.
  • Real-time adaptive control becomes feasible for open quantum systems whose size or noise complexity renders repeated optimization intractable.

Where Pith is reading between the lines

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

  • If the latent dimension remains small when system size grows, the same architecture could scale to larger many-body systems where conventional optimal control fails.
  • The learned manifold might be inspected post-training to extract simplified effective models of the controlled open dynamics.
  • Coupling the trained network to experimental feedback loops could enable online recalibration when the physical noise deviates from the training distribution.

Load-bearing premise

A low-dimensional latent manifold extracted from training trajectories is sufficient to represent the dynamics needed for control optimization across the tested noise levels and system sizes.

What would settle it

A concrete test would be to generate new initial states and time-varying noise parameters outside the training distribution, run the model once, and compare its predicted trajectories against exact numerical integration; large systematic deviations would falsify the claim that the manifold captures the relevant dynamics.

Figures

Figures reproduced from arXiv: 2606.27907 by Feng-Hua Ren, Jun-Dong Zhong, Zhao-Ming Wang, Zong-Yuan Ge.

Figure 1
Figure 1. Figure 1: FIG. 1: End-to-end quantum control framework based on la [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Adiabatic speedup in a two-level open quantum [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Parameter generalization and stability of the pro [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Adiabatic fidelity without control as a function of the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: State transfer in a one-dimensional XY spin chain. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Schematic representations of RNNs and LSTM archi [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Training and evaluation loss evolution for a two-level [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9: Adiabatic fidelity evolution for a two-level open quantum system under time-varying noise ( [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: Generalization performance with respect to environ [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11: Generalization of the proposed framework with respect to initial states and driving field in adiabatic state preparation [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
read the original abstract

Traditional quantum control relies on an iterative "simulate-then-optimize" paradigm, where dynamics simulation and control design are decoupled, leading to substantial computational overhead and limited scalability, particularly in noisy environments. Here, we propose an end-to-end quantum control framework based on long short-term memory, in which system dynamics and control strategies are learned jointly in a low dimensional latent manifold. The model directly maps initial states and environmental parameters to both dynamical trajectories and optimized control pulse in a single forward pass. The framework is validated on adiabatic speedup in a two-level system and state transfer in a one-dimensional spin chain under noise, achieving accurate dynamical prediction and control optimization. It improves the fidelity for both tasks and significantly reduces the optimization cost by three orders of magnitude compared with conventional iterative methods, while exhibiting strong generalization to multi-parameter, time-varying noise, as well as to different initial states and driving fields. Our work introduces a data-driven control paradigm based on latent manifold learning, reducing the computational bottleneck of iterative optimization and enabling real-time adaptive control of complex open quantum systems.

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 introduces an end-to-end LSTM-based framework for quantum control that jointly learns open-system dynamics and optimized control pulses within a low-dimensional latent dynamical manifold. The model maps initial states and environmental parameters directly to trajectories and control fields in one forward pass. It is validated on adiabatic speedup for a two-level system and state transfer in a noisy 1D spin chain, claiming higher fidelity than conventional methods together with a three-order-of-magnitude reduction in optimization cost and generalization across multi-parameter time-varying noise, different initial states, and driving fields.

Significance. If the performance claims hold under rigorous verification, the work could enable real-time adaptive control for open quantum systems by eliminating iterative simulate-then-optimize loops. The joint learning of dynamics and control on a latent manifold represents a data-driven alternative to traditional optimal-control techniques. Credit is due for the reported generalization tests across noise regimes and system sizes, which go beyond single-task demonstrations common in the literature.

major comments (2)
  1. [Abstract, §5] Abstract and §5 (results): The central performance claims of improved fidelity and 1000× speedup are stated without reported error bars, training-set sizes, number of independent runs, or explicit baseline comparisons (e.g., GRAPE or Krotov iterations with identical noise realizations). This absence makes it impossible to judge whether the gains are statistically robust or sensitive to post-hoc data selection.
  2. [§4.2, §5.3] §4.2 (latent-manifold construction) and §5.3 (generalization tests): The claim that the learned manifold faithfully captures relevant open-system dynamics rests on training trajectories alone. No explicit check is provided that the reduced representation preserves positivity or phase information for noise correlations outside the training distribution (e.g., non-Markovian kernels with longer memory times), which is load-bearing for the optimality of the inferred control pulses.
minor comments (2)
  1. [§3] Notation for the latent state dimension and the precise form of the loss function (reconstruction plus control) should be introduced earlier and used consistently.
  2. [Figures 3–5] Figure captions should state the number of test trajectories and the precise noise-parameter ranges used for each panel.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive evaluation of the work's potential. We respond to each major comment below with clarifications and commitments to revisions that address the raised concerns directly.

read point-by-point responses

  1. Referee: [Abstract, §5] Abstract and §5 (results): The central performance claims of improved fidelity and 1000× speedup are stated without reported error bars, training-set sizes, number of independent runs, or explicit baseline comparisons (e.g., GRAPE or Krotov iterations with identical noise realizations). This absence makes it impossible to judge whether the gains are statistically robust or sensitive to post-hoc data selection.

    Authors: We agree that the submitted manuscript does not report error bars, the number of independent runs, training-set sizes, or side-by-side baseline comparisons with identical noise realizations in the abstract and §5. This omission limits evaluation of statistical robustness. In the revised manuscript we will add standard deviations computed over multiple independent training runs (minimum of five seeds), explicitly state the training-set sizes used for each task, and include direct fidelity and runtime comparisons against GRAPE and Krotov executed on the same noise realizations. revision: yes

  2. Referee: [§4.2, §5.3] §4.2 (latent-manifold construction) and §5.3 (generalization tests): The claim that the learned manifold faithfully captures relevant open-system dynamics rests on training trajectories alone. No explicit check is provided that the reduced representation preserves positivity or phase information for noise correlations outside the training distribution (e.g., non-Markovian kernels with longer memory times), which is load-bearing for the optimality of the inferred control pulses.

    Authors: The referee correctly observes that explicit verification of positivity and phase preservation for out-of-distribution noise (e.g., longer-memory non-Markovian kernels) is absent; the generalization results in §5.3 report performance metrics but do not directly test these physical invariants. We will add the requested checks in the revision by reconstructing density matrices from the latent states for extrapolated noise kernels, confirming that eigenvalues remain non-negative and that relative phases are consistent with the original open-system evolution. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper proposes a data-driven LSTM architecture that learns a joint mapping from states/parameters to trajectories and controls on a latent manifold, with performance validated empirically on specific tasks. No load-bearing mathematical derivation, equation, or self-citation reduces by construction to its own inputs; the framework is an empirical ML method whose outputs are not tautologically equivalent to fitted training quantities in the sense of the enumerated patterns. Claims rest on simulation benchmarks rather than self-referential definitions.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields limited visibility into training details; the central claim rests on the existence of a faithful low-dimensional latent representation learned from data.

free parameters (1)
  • LSTM network weights
    All model parameters are fitted during training on simulated trajectories; exact count and regularization choices unknown from abstract.

pith-pipeline@v0.9.1-grok · 5719 in / 1174 out tokens · 36442 ms · 2026-06-29T04:27:47.004828+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

75 extracted references · 3 canonical work pages · 2 internal anchors

  1. [1]

    Bluvstein, S

    D. Bluvstein, S. J. Evered, A. A. Geim, S. H. Li, H. Zhou, T. Manovitz, S. Ebadi, M. Cain, M. Kalinowski, D. Hangleiter, et al., Nature626, 58–65 (2023), ISSN 1476-4687

    2023

  2. [2]

    Giovannetti, S

    V. Giovannetti, S. Lloyd, and L. Maccone, Physical Re- view Letters96(2006), ISSN 1079-7114

    2006

  3. [3]

    Altman, K

    E. Altman, K. R. Brown, G. Carleo, L. D. Carr, E. Dem- ler, C. Chin, B. DeMarco, S. E. Economou, M. A. Eriks- son, K.-M. C. Fu, et al., PRX Quantum2(2021), ISSN 2691-3399

    2021

  4. [4]

    P. M. Poggi, G. De Chiara, S. Campbell, and A. Kiely, Physical Review Letters132(2024), ISSN 1079-7114

    2024

  5. [5]

    L. Ding, J. Fan, and X. Qiu, npj Quantum Information 12(2025), ISSN 2056-6387

    2025

  6. [6]

    Gao, N.-J

    J.-L. Gao, N.-J. Cui, H.-Y. Ma, and Z.-M. Wang, Physi- cal Review A113, 042408 (2026)

    2026

  7. [7]

    Shi, J.-T

    Z.-C. Shi, J.-T. Ding, Y.-H. Chen, J. Song, Y. Xia, X. Yi, and F. Nori, Physical Review Applied21(2024), ISSN 2331-7019

    2024

  8. [8]

    Zhang, T

    J. Zhang, T. H. Kyaw, S. Filipp, L.-C. Kwek, E. Sj¨ oqvist, and D. Tong, Physics Reports1027, 1 (2023), ISSN 0370- 1573, geometric and holonomic quantum computation

    2023

  9. [9]

    Y.-J. Hai, Y. Song, J. Li, J. Zeng, and X.-H. Deng, Phys. Rev. Appl.23, 054002 (2025)

    2025

  10. [10]

    Jing, L.-A

    J. Jing, L.-A. Wu, M. Byrd, J. Q. You, T. Yu, and Z.-M. Wang, Phys. Rev. Lett.114, 190502 (2015)

    2015

  11. [11]

    Z.-M. Wang, M. S. Byrd, J. Jing, and L.-A. Wu, Physical Review A97(2018), ISSN 2469-9934

    2018

  12. [12]

    Z.-M. Wang, M. S. Sarandy, and L.-A. Wu, Phys. Rev. A102, 022601 (2020)

    2020

  13. [13]

    Wang, F.-H

    Z.-M. Wang, F.-H. Ren, D.-W. Luo, Z.-Y. Yan, and L.-A. Wu, Phys. Rev. A102, 042406 (2020)

    2020

  14. [14]

    Xie, F.-H

    Y.-Y. Xie, F.-H. Ren, R.-H. He, A. Ablimit, and Z.-M. Wang, Physical Review A106(2022), ISSN 2469-9934

    2022

  15. [15]

    Khaneja, T

    N. Khaneja, T. Reiss, C. Kehlet, T. Schulte-Herbr¨ uggen, and S. J. Glaser, Journal of Magnetic Resonance172, 296–305 (2005), ISSN 1090-7807

    2005

  16. [16]

    Caneva, T

    T. Caneva, T. Calarco, and S. Montangero, Physical Re- view A84(2011), ISSN 1094-1622

    2011

  17. [17]

    F. F. Floether, P. d. Fouquieres, and S. G. Schirmer, New Journal of Physics14, 073023 (2012), ISSN 1367-2630

    2012

  18. [18]

    M. E. F. Fernandes, F. F. Fanchini, E. F. de Lima, and L. K. Castelano, Journal of Physics A: Mathematical and Theoretical56, 495303 (2023), ISSN 1751-8121

    2023

  19. [19]

    P. J. Werbos, Proceedings of the IEEE78, 1550 (1990)

    1990

  20. [20]

    Li, Applied and Computational Engineering133, 165–173 (2025), ISSN 2755-273X

    Z. Li, Applied and Computational Engineering133, 165–173 (2025), ISSN 2755-273X

    2025

  21. [21]

    L. J. Bond, B. Gerritsen, J. Min´ aˇ r, J. T. Young, J. Schachenmayer, and A. Safavi-Naini, The Journal of 6 Chemical Physics161(2024)

    2024

  22. [22]

    Guo and X

    Y. Guo and X. Gao, Journal of Chemical Theory and Computation21, 8650 (2025)

    2025

  23. [23]

    Banchi, E

    L. Banchi, E. Grant, A. Rocchetto, and S. Severini, New Journal of Physics20, 123030 (2018), ISSN 1367-2630

    2018

  24. [24]

    Wang and Z.-M

    Z.-W. Wang and Z.-M. Wang,Time series prediction of open quantum system dynamics(2024)

    2024

  25. [25]

    D. Luo, Z. Chen, J. Carrasquilla, and B. K. Clark, Phys- ical Review Letters128(2022), ISSN 1079-7114

    2022

  26. [26]

    I. D. Mienye, T. G. Swart, and G. Obaido, Information 15, 517 (2024), ISSN 2078-2489

    2024

  27. [27]

    Z. An, J. Wu, Z. Lin, X. Yang, K. Li, and B. Zeng, Physical Review Letters134, 120202 (2025)

    2025

  28. [28]

    Lusch, J

    B. Lusch, J. N. Kutz, and S. L. Brunton, Nature com- munications9, 4950 (2018)

    2018

  29. [29]

    Rubanova, R

    Y. Rubanova, R. T. Chen, and D. K. Duvenaud, Ad- vances in neural information processing systems32 (2019)

    2019

  30. [30]

    R. G. Krishnan, U. Shalit, and D. Sontag, arXiv preprint arXiv:1511.05121 (2015)

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  31. [31]

    K. Lin, J. Peng, F. L. Gu, and Z. Lan, The Journal of Physical Chemistry Letters12, 10225–10234 (2021), ISSN 1948-7185

    2021

  32. [32]

    Mohseni, T

    N. Mohseni, T. F¨ osel, L. Guo, C. Navarrete-Benlloch, and F. Marquardt, Quantum6, 714 (2022), ISSN 2521- 327X

    2022

  33. [33]

    Zhong and Z.-M

    J.-D. Zhong and Z.-M. Wang, Physical Review A113, 062442 (2026)

    2026

  34. [34]

    Norambuena, M

    A. Norambuena, M. Mattheakis, F. J. Gonz´ alez, and R. Coto, Physical Review Letters132(2024), ISSN 1079- 7114

    2024

  35. [35]

    Manzano, AIP Advances10(2020), ISSN 2158-3226

    D. Manzano, AIP Advances10(2020), ISSN 2158-3226

    2020

  36. [36]

    Yu, Physical Review A69(2004), ISSN 1094-1622

    T. Yu, Physical Review A69(2004), ISSN 1094-1622

    2004

  37. [37]

    Wang, F.-H

    Z.-M. Wang, F.-H. Ren, D.-W. Luo, Z.-Y. Yan, and L.-A. Wu, Journal of Physics A: Mathematical and Theoretical 54, 155303 (2021), ISSN 1751-8121

    2021

  38. [38]

    W. T. Strunz, L. Di´ osi, and N. Gisin, Physical review letters82, 1801 (1999)

    1999

  39. [39]

    J. L. Elman, Cognitive science14, 179 (1990)

    1990

  40. [40]

    Hochreiter, Neural Computation MIT-Press (1997)

    S. Hochreiter, Neural Computation MIT-Press (1997)

    1997

  41. [41]

    Greff, R

    K. Greff, R. K. Srivastava, J. Koutn´ ık, B. R. Steune- brink, and J. Schmidhuber, IEEE transactions on neural networks and learning systems28, 2222 (2016)

    2016

  42. [42]

    J. P. Kestner, X. Wang, L. S. Bishop, E. Barnes, and S. Das Sarma, Phys. Rev. Lett.110, 140502 (2013)

    2013

  43. [43]

    X. Rong, J. Geng, Z. Wang, Q. Zhang, C. Ju, F. Shi, C.- K. Duan, and J. Du, Physical review letters112, 050503 (2014)

    2014

  44. [44]

    T. Xie, Z. Zhao, S. Xu, et al., Physical Review Letters 130, 030601 (2023)

    2023

  45. [45]

    Hornik, M

    K. Hornik, M. Stinchcombe, and H. White, Neural net- works2, 359 (1989)

    1989

  46. [46]

    R. C. Staudemeyer and E. R. Morris,Understanding lstm – a tutorial into long short-term memory recurrent neural networks(2019)

    2019

  47. [47]

    Krichen and A

    M. Krichen and A. Mihoub, AI6, 215 (2025), ISSN 2673- 2688

    2025

  48. [48]

    R. T. Chen, Y. Rubanova, J. Bettencourt, and D. K. Du- venaud, Advances in neural information processing sys- tems31(2018)

    2018

  49. [49]

    Universal Differential Equations for Scientific Machine Learning

    C. Rackauckas, Y. Ma, J. Martensen, C. Warner, K. Zubov, R. Supekar, D. Skinner, A. Ramadhan, and A. Edelman, arXiv preprint arXiv:2001.04385 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2001

  50. [50]

    Qui˜ nonero-Candela, M

    J. Qui˜ nonero-Candela, M. Sugiyama, A. Schwaighofer, and N. D. Lawrence,Dataset shift in machine learning (Mit Press, 2008)

    2008

  51. [51]

    Farhi, J

    E. Farhi, J. Goldstone, S. Gutmann, J. Lapan, A. Lund- gren, and D. Preda, Science292, 472–475 (2001), ISSN 1095-9203

    2001

  52. [52]

    J. Lin, Z. Y. Lai, and X. Li, Physical Review A101 (2020), ISSN 2469-9934

    2020

  53. [53]

    L.-A. Wu, M. S. Byrd, and D. A. Lidar, Physical Review Letters89(2002), ISSN 1079-7114

    2002

  54. [54]

    M. A. Nielsen and I. L. Chuang,Quantum computation and quantum information(Cambridge university press, 2010)

    2010

  55. [55]

    Yan, Chinese Journal of Chemical Physics30, 277–286 (2017), ISSN 2327-2244

    Y.-a. Yan, Chinese Journal of Chemical Physics30, 277–286 (2017), ISSN 2327-2244

    2017

  56. [56]

    Rivas, S

    ´A. Rivas, S. F. Huelga, and M. B. Plenio, Reports on Progress in Physics77, 094001 (2014)

    2014

  57. [57]

    S. J. Glaser, U. Boscain, T. Calarco, C. P. Koch, W. K¨ ockenberger, R. Kosloff, I. Kuprov, B. Luy, S. Schirmer, T. Schulte-Herbr¨ uggen, et al., The Euro- pean Physical Journal D69, 279 (2015)

    2015

  58. [58]

    C. Brif, R. Chakrabarti, and H. Rabitz, New Journal of Physics12, 075008 (2010)

    2010

  59. [59]

    Dong and I

    D. Dong and I. R. Petersen,Learning and robust control in quantum technology(Springer, 2023)

    2023

  60. [60]

    Y. Lee, I. R. Petersen, and D. Dong, arXiv preprint arXiv:2402.07396 (2024)

    work page arXiv 2024

  61. [61]

    Wang and T.-Z

    Z.-M. Wang and T.-Z. Chen, Physical Review Research 6, 043195 (2024)

    2024

  62. [62]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, Nature Reviews Physics3, 422 (2021)

    2021

  63. [63]

    Christandl, N

    M. Christandl, N. Datta, A. Ekert, and A. J. Landahl, Physical review letters92, 187902 (2004)

    2004

  64. [64]

    Christandl, N

    M. Christandl, N. Datta, T. C. Dorlas, A. Ekert, A. Kay, and A. J. Landahl, Physical Review A—Atomic, Molec- ular, and Optical Physics71, 032312 (2005)

    2005

  65. [65]

    Ren, Z.-M

    F.-H. Ren, Z.-M. Wang, and Y.-J. Gu, Quantum Infor- mation Processing18, 1 (2019)

    2019

  66. [66]

    Breuer and F

    H.-P. Breuer and F. Petruccione,The theory of open quantum systems(OUP Oxford, 2002)

    2002

  67. [67]

    Chen and T

    Y. Chen and T. Yu,Non-Markovian Dynamics of Qubit Systems: Quantum-State Diffusion Equations Versus Master Equations(Springer International Publishing, 2015), p. 609–634, ISBN 9783319253404

    2015

  68. [68]

    L.-A. Wu, G. Kurizki, and P. Brumer, Phys. Rev. Lett. 102, 080405 (2009)

    2009

  69. [69]

    Wang, L.-A

    Z.-M. Wang, L.-A. Wu, J. Jing, B. Shao, and T. Yu, Phys. Rev. A86, 032303 (2012)

    2012

  70. [70]

    Bengio, J

    Y. Bengio, J. Louradour, R. Collobert, and J. Weston, in Proceedings of the 26th annual international conference on machine learning(2009), pp. 41–48

    2009

  71. [71]

    A. G. Baydin, B. A. Pearlmutter, A. A. Radul, and J. M. Siskind, Journal of machine learning research18, 1 (2018)

    2018

  72. [72]

    M. Y. Niu, S. Boixo, V. N. Smelyanskiy, and H. Neven, npj Quantum Information5, 33 (2019)

    2019

  73. [73]

    END TO END LEARNING OF QUANTUM CONTROL ON LA TENT DYNAMICAL MANIFOLD

    K. Champion, B. Lusch, J. N. Kutz, and S. L. Brunton, Proceedings of the National Academy of Sciences116, 22445 (2019). 7 SUPPLEMENT AL MA TERIAL: “END TO END LEARNING OF QUANTUM CONTROL ON LA TENT DYNAMICAL MANIFOLD” S1: NON-MARKOVIAN QUANTUM MASTER EQUA TIONS Consider the dynamics of an open quantum system coupled to non-Markovian bosonic baths [38, 66]...

    2019

  74. [74]

    withO(0) taken as the true initial observable

    is recursively fed back as part of the next inputx t+1. withO(0) taken as the true initial observable. Compared with standard RNNs, LSTM introduces gating mecha- nisms and an additive cell-state update, enabling stable long-sequence modeling by mitigating vanishing and ex- ploding gradients [40]. Its capability originates from its gated memory mechanism, ...

  75. [75]

    remain low, confirming accurate dynamical prediction over the entire evolution. These results further indicate that the Initial-State Encoder effectively incorporates in- formation from the initial-state into the latent initial con- dition (h 0, c0), while the network input and the man- ifold’s generalization constraints enable consistent roll- out of dyn...