Pith. sign in

REVIEW 2 major objections 5 minor 32 references

A multilayer perceptron, trained on STIRAP transfer efficiencies measured under eight driving and detuning configurations, can reconstruct the gauge-invariant plaquette phase of a three-level Δ system across the full 2π interval.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 00:19 UTC pith:6HSTI6JV

load-bearing objection A clean simulation-level proof of principle for estimating the plaquette phase from STIRAP efficiencies; the 'sensing' language runs ahead of the evidence, and the sign-degeneracy objection doesn't survive contact with the paper. the 2 major comments →

arxiv 2607.15040 v1 pith:6HSTI6JV submitted 2026-07-16 quant-ph

Machine-Learning-Empowered Quantum Sensing of the Plaquette Phase in a Three-Level Delta System

Lorenzo Vitale , Shreyasi Mukherjee , Dario Fasone , Enrico Martello , Elisabetta Paladino , Luigi Giannelli , Giuseppe Falci This is my paper
classification quant-ph
keywords plaquette phasethree-level Δ systemSTIRAPcoherent population trappingmachine learningmultilayer perceptronquantum sensingsynthetic gauge fields
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper proposes a method to measure the gauge-invariant plaquette phase of a closed-loop three-level system—a quantity that is not directly observable—by turning the known phase-dependent degradation of STIRAP population transfer into a sensing signal. The central claim is that a multilayer perceptron, trained on simulated STIRAP efficiencies collected under eight different driving and detuning conditions, can accurately reconstruct the phase across the full 2π range. If true, this gives a cheap, measurement-tolerant way to sense an effective magnetic flux in engineered quantum systems, with applications to synthetic gauge fields and quantum metrology. The insight is that an imperfection—the breakdown of coherent population trapping—carries precisely the information one wants.

Core claim

The paper demonstrates that the plaquette phase φ, a gauge-invariant quantity that cannot be directly measured, is encoded in the transfer efficiency of STIRAP. When the detuning is set to the value that would ensure perfect trapping at φ=0, any nonzero φ causes leakage from the dark manifold proportional to sin φ, so the final population of the target state becomes a phase-dependent observable. The authors construct an eight-dimensional fingerprint of efficiencies from three driving conditions and three detuning schemes, and show that a multilayer perceptron trained on noiseless simulated data predicts φ across the full interval with small error. The MLP outputs (cos φ, sin φ), avoiding the

What carries the argument

The gauge-invariant plaquette phase φ = φ0 + φs − φp, which remains after all local gauge transformations are exhausted; the generalized dark-state (trapped-state) condition that defines the detuning baseline; the dark state |D⟩ = cosθ|0⟩ − sinθ|1⟩; and the fact that for φ ≠ 0 the dressed states acquire a component along |D⟩ proportional to sin φ, which produces phase-dependent leakage. The sensing protocol uses eight STIRAP efficiencies (three driving configurations × detuning settings) as features, and a two-output MLP trained with mean-squared error on the circular encoding.

Load-bearing premise

The eight measured transfer efficiencies form a well-conditioned, effectively invertible fingerprint of the plaquette phase, so a network trained on ideal noiseless simulations will infer the phase from real measurements made with noise and experimental imperfections.

What would settle it

Sweep the plaquette phase on a controllable platform (or simulate the full Lindblad master equation with realistic noise) and record the eight efficiencies; if two distinct phases in [−π,π] yield fingerprint vectors that are statistically indistinguishable at the experimental noise level, or if the network's prediction error grows sharply once noise is added to the ideal efficiencies, then the central sensing claim fails. A numerical check: compute the Jacobian of the eight-efficiency map with respect to φ; if its singular values collapse toward zero anywhere, the map is locally non-invertible

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • A single final-population measurement per configuration replaces full state tomography, making the phase estimate cheap and experimentally accessible.
  • The phase-dependent degradation of STIRAP, normally a drawback, becomes the physical resource that carries the sensing signal.
  • Because the method relies only on the closed-loop geometry, it transfers to any engineered platform—superconducting circuits, NV centers, quantum dots—with the same fingerprint structure.
  • The circular (cos φ, sin φ) output means the estimator works uniformly across the 2π boundary, so no phase discontinuity artifacts appear.

Where Pith is reading between the lines

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

  • The paper only tests noiseless synthetic data; the immediate next experiment is to add measurement noise and parameter fluctuations. If the eight-efficiency mapping is not robust, the authors will need to augment the training set or the feature vector.
  • The choice of eight features is heuristic (three driving conditions, three detunings, two matched configurations). A cheaper protocol with fewer features may be possible; the paper lists reducing the number of input features as future work.
  • In a lattice of many triangular plaquettes, the same method could serve as a local magnetometer for synthetic gauge fields: measuring each plaquette's phase map would reconstruct the effective magnetic-field texture.
  • The network's generalization is only demonstrated within the same parameter range used for training (fixed pulse width, delay, and amplitude ratio). Whether the estimator is robust to changes in pulse shape or timing is an untested extension.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes a machine-learning approach to estimate the plaquette phase φ in a closed-loop three-level Δ system from STIRAP population transfer efficiencies. The Hamiltonian (2) contains a residual complex coupling e^{±iφ}; Sec. 2.3 shows that a non-zero φ breaks coherent population trapping, producing a phase-dependent transfer efficiency. Sec. 3 describes an eight-efficiency feature vector built from three driving conditions and several detuning protocols, and an MLP with circular output (cos φ, sin φ). Sec. 4 reports low test error on a synthetic dataset of 2000 phases. The authors conclude that the phase can be accurately reconstructed from experimentally accessible observables.

Significance. The analytical treatment of the dark-state breakdown is standard and the data-generation pipeline is transparent, including a held-out test set and early stopping; these are strengths. However, the central sensing claim is invalidated by an exact symmetry: all eight input efficiencies are invariant under φ→-φ while the target is not. Thus the reported MLP performance cannot be reproduced as stated, and the proposed method does not sense the signed plaquette phase. The practical significance of the work therefore hinges on a symmetry obstruction that is not addressed; the paper would need a fundamentally different observable or a reduced target (e.g., |φ|) to be salvageable.

major comments (2)
  1. [Secs. 3.3 and 4, Eq. (2)] For the Hamiltonian in Eq. (2), H(−φ)=H(φ)^*. Since the pulses in Eq. (10) are real, the detunings are real, and the STIRAP initial state is |0⟩, the propagator satisfies U(−φ)=U(φ)^*. Every feature is a final population η=|<1|ψ(t_f)>|² (Eq. 12), so η_k(−φ)=η_k(φ) for each of the eight configurations. The input vector therefore satisfies x(−φ)=x(φ) while the target (cos φ, sin φ) in Eq. (13) is not invariant under φ→−φ. A deterministic MLP cannot separate φ and −φ from identical inputs. The low test error in Fig. 4 is inconsistent with the stated data-generation protocol unless the symmetry is broken by an unstated step. This is a load-bearing flaw, not a noise issue: it persists in the ideal noiseless limit.
  2. [Abstract; Sec. 5] The abstract claims estimation from 'experimentally accessible observables' and the conclusions frame the result as a quantum-sensing protocol. However, the entire study uses noiseless synthetic data with no detector noise, no parameter fluctuations, no error bars, and no comparison with a direct estimator or baseline. The only generalization demonstrated is interpolation within the same model distribution. This overstates the experimental scope and should be corrected even if the symmetry issue were resolved.
minor comments (5)
  1. [Sec. 5, first paragraph] 'three-level ∆ system with.' contains a dangling 'with'.
  2. [Abstract] 'STImulated' should be 'Stimulated'.
  3. [Eq. (8)] The notation eδ_p is undefined; it presumably denotes the shifted detuning defined in Eq. (7).
  4. [Sec. 3.3, Table 2] The definitions of δ_pump and δ_Stokes are described only in prose; the table would be clearer if the matched detuning formulas were written explicitly.
  5. [Sec. 4 and Fig. 4] The conversion from the two output components to a predicted angle is not specified; the convention used to report 'predicted phase' should be stated.

Circularity Check

0 steps flagged

No circular derivation; physics is from external Ref. [15], ML test is held-out interpolation, self-citations are motivational only.

full rationale

The derivation chain is self-contained. The trapping condition, dark state, and dressed-state leakage (Eqs. (4)-(8)) are taken from Ref. [15], an external reference, and the MLP is trained on numerically integrated Schrödinger dynamics of Eq. (2) with held-out test phases; the target y=(cos φ, sin φ) is not used as an input and no fitted parameter is renamed as a prediction. The only self-citations [23,24] appear as motivation ('we take inspiration') and for a normalization convention; they carry no load in the sensing argument. The paper itself flags the inverse problem as 'generally non-invertible' (Sec. 1) and defers noise robustness to future work (Sec. 5); these are limitations of the protocol, not circularity. Any φ→−φ identifiability issue would be a physical correctness concern rather than a definitional/fitted-input circularity and is not counted here.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

All training data are generated from the same Hamiltonian that defines the target phase, and the paper provides no experimental noise model or invertibility proof; these are the main uncharged premises.

free parameters (4)
  • Pulse width T = ΩT = 15
    Chosen simulation parameter; the phase-efficiency mapping and thus the sensing fidelity depend on the adiabaticity of the STIRAP pulses.
  • Pulse delay τ = 0.7 T
    Chosen for counterintuitive STIRAP sequence; affects transfer efficiency and the shape of the phase-dependent signal.
  • Detuning δ_p = 0
    Set to zero; part of the driving conditions used to generate the eight input features.
  • Pulse amplitude ratios = Ω_p/Ω_s ∈ {1, 2, 1/2}
    Chosen driving configurations; essential for disambiguating the phase since a single efficiency is non-invertible.
axioms (5)
  • domain assumption Rotating-wave three-level Δ Hamiltonian with complex couplings, Eqs. (1)-(2)
    All training data are generated from this model; if the model does not describe a real device, the sensing claim fails.
  • domain assumption The trapped-state/dressed-state analysis from Ref. [15] (Sec. 2.3) remains valid under the operational detuning of Eq. (9)
    Used to predict phase-dependent leakage that the ML exploits.
  • ad hoc to paper The mapping from phase φ to the 8-dimensional efficiency vector is well-posed (unique and stable enough to invert)
    No proof; the paper relies on empirical training success to justify the chosen feature set (Sec. 3.3, Table 2).
  • standard math An MLP can approximate this mapping and generalize to unseen phases
    Universal approximation / supervised learning; standard but not proven for this task.
  • domain assumption Measured transfer efficiencies are noiseless and exact
    Synthetic data are generated by solving the Schrödinger equation with no detection noise; real efficiencies will have noise, and robustness is deferred to future work (Sec. 5).

pith-pipeline@v1.3.0-alltime-deepseek · 9018 in / 11933 out tokens · 116652 ms · 2026-08-02T00:19:08.976909+00:00 · methodology

0 comments
read the original abstract

We propose a machine-learning-empowered approach to the quantum sensing of the plaquette phase, a gauge-invariant quantity arising in three-level $\Delta$ systems. This phase profoundly affects the system dynamics, breaking coherent population trapping and inducing a non-trivial phase dependence of the dynamics. We demonstrate that a multi-layer perceptron (MLP), trained in a supervised-learning framework, can accurately estimate the plaquette phase from STImulated Raman Adiabatic Passage (STIRAP) population transfer efficiencies measured under different driving conditions, which provide experimentally accessible observables. Our results highlight how the combination of coherent control and machine learning (ML) enables effective phase identification in closed-loop quantum systems, opening new perspectives for quantum technologies, specifically quantum sensing applications including synthetic gauge fields.

Figures

Figures reproduced from arXiv: 2607.15040 by Dario Fasone, Elisabetta Paladino, Enrico Martello, Giuseppe Falci, Lorenzo Vitale, Luigi Giannelli, Shreyasi Mukherjee.

Figure 1
Figure 1. Figure 1: Three-level system in a ∆ configuration. sensing of the plaquette phase in a three-level ∆ system. By exploiting the phase-dependent degradation of STIRAP transfer efficiency under different driving conditions, we train a NN to reconstruct the phase from experimen￾tally accessible observables. This approach highlights how coherent control protocols, when combined with data-driven methods, can be repurposed… view at source ↗
Figure 2
Figure 2. Figure 2: Time- and phase-resolved population of the target state |1⟩ during the STIRAP protocol, obtained under the operational two-photon detuning condition of Eq. (9). The Gaussian pulses given by Eq. (10) are taken with equal peak amplitudes, Ωmax p = Ωmax s = Ω. Simulation parameters: ΩT = 15, τ = 0.7 T, and δp = 0. specifically a MLP. The network is trained on a labeled dataset, {(xi , ˆyi)}i=1,...,N , where e… view at source ↗
Figure 3
Figure 3. Figure 3: Training progress. The value of the cost function, given by Eq. (14), for the training sets (solid blue line) and the validation sets (solid orange line) vs. the number of epochs of training. quently optimized through gradient-based minimization of the loss function Eq. (14). The training process of the model is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance of the neural network on the test dataset. Left: predicted phase versus the corresponding target value (the true value). Right: prediction accuracy as a function of the test sample index. 5 Conclusions In this work, we have proposed a ML–assisted approach to the sensing of the plaquette phase in a three-level ∆ system with. By analyzing the impact of the gauge-invariant phase on the system dyna… view at source ↗

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

32 extracted references · 6 canonical work pages

  1. [1]

    The Quantum Technologies Roadmap: A European Community View

    A. Ac ´ ın et al. “The Quantum Technologies Roadmap: A European Community View”. In:New Journal of Physics20 (2018), p. 080201. issn: 1367-2630.doi:10.1088/1367-2630/aad1ea

    doi:10.1088/1367-2630/aad1ea 2018

  2. [2]

    Quantum Technology: The Sec- ond Quantum Revolution

    J. P. Dowling and G. J. Milburn. “Quantum Technology: The Sec- ond Quantum Revolution”. In:Philosophical Transactions of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences361 (2003). Ed. by A. G. J. MacFarlane, pp. 1655–1674.issn: 1364-503X, 1471-2962.doi:10.1098/rsta.2003.1227

    arXiv 2003

  3. [3]

    Quantum Sensing

    C. L. Degen, F. Reinhard, and P. Cappellaro. “Quantum Sensing”. In:Reviews of Modern Physics89 (2017), p. 035002.issn: 0034-6861, 1539-0756.doi:10.1103/RevModPhys.89.035002

    doi:10.1103/revmodphys.89.035002 2017

  4. [4]

    Coherent Manipulations of Atoms Using Laser Light

    B. Shore. “Coherent Manipulations of Atoms Using Laser Light”. In: Acta Physica Slovaca. Reviews and Tutorials58.3 (2008).issn: 1336- 040X, 0323-0465.doi:10.2478/v10155-010-0090-z

    work page doi:10.2478/v10155-010-0090-z 2008

  5. [5]

    Nonabsorbing Atomic Coherences by Coherent Two-Photon Transitions in a Three-Level Optical Pumping

    E. Arimondo and G. Orriols. “Nonabsorbing Atomic Coherences by Coherent Two-Photon Transitions in a Three-Level Optical Pumping”. In:Lettere al Nuovo Cimento (1971-1985)17.10 (1976), pp. 333–338. issn: 1827-613X.doi:10.1007/BF02746514

    work page doi:10.1007/bf02746514 1971

  6. [6]

    Coherent Trapping of Atomic Populations

    H. R. Gray, R. M. Whitley, and C. R. Stroud. “Coherent Trapping of Atomic Populations”. In:Optics Letters3.6 (1978), pp. 218–220.issn: 1539-4794.doi:10.1364/OL.3.000218. 13

    doi:10.1364/ol.3.000218 1978

  7. [7]

    V coherent population trapping in laser spectroscopy

    E. Arimondo. “V coherent population trapping in laser spectroscopy”. In:Progress in optics35 (1996), pp. 257–354.doi:10.1016/S0079- 6638(08)70531-6

    doi:10.1016/s0079- 1996

  8. [8]

    Coherent Population Transfer among Quantum States of Atoms and Molecules

    K. Bergmann, H. Theuer, and B. W. Shore. “Coherent Population Transfer among Quantum States of Atoms and Molecules”. In:Re- views of Modern Physics70.3 (1998), pp. 1003–1025.doi:10.1103/ RevModPhys.70.1003

    1998

  9. [9]

    Stimulated Raman Adiabatic Passage in Physics, Chemistry, and Beyond

    N. V. Vitanov et al. “Stimulated Raman Adiabatic Passage in Physics, Chemistry, and Beyond”. In:Reviews of Modern Physics89.1 (2017), p. 015006.doi:10.1103/RevModPhys.89.015006

    doi:10.1103/revmodphys.89.015006 2017

  10. [10]

    Stimulated Raman Adiabatic Passage in a Three- Level Superconducting Circuit

    K. S. Kumar et al. “Stimulated Raman Adiabatic Passage in a Three- Level Superconducting Circuit”. In:Nature Communications7.1 (2016), p. 10628.issn: 2041-1723.doi:10.1038/ncomms10628

    doi:10.1038/ncomms10628 2016

  11. [11]

    Advanced Control with a Cooper- pair Box: Stimulated Raman Adiabatic Passage and Fock-state Gen- eration in a Nanomechanical Resonator

    J. Siewert, T. Brandes, and G. Falci. “Advanced Control with a Cooper- pair Box: Stimulated Raman Adiabatic Passage and Fock-state Gen- eration in a Nanomechanical Resonator”. In:Physical Review B79.2 (2009), p. 024504.doi:10.1103/PhysRevB.79.024504

    doi:10.1103/physrevb.79.024504 2009

  12. [12]

    Design of a Lambda System for Population Transfer in Superconducting Nanocircuits

    G. Falci et al. “Design of a Lambda System for Population Transfer in Superconducting Nanocircuits”. In:Physical Review B87.21 (2013), p. 214515.doi:10.1103/PhysRevB.87.214515

    doi:10.1103/physrevb.87.214515 2013

  13. [13]

    Advances in Quantum Control of Three-Level Super- conducting Circuit Architectures

    G. Falci et al. “Advances in Quantum Control of Three-Level Super- conducting Circuit Architectures”. In:Fortschritte der Physik65.6-8 (2017), p. 1600077.issn: 1521-3978.doi:10.1002/prop.201600077

    doi:10.1002/prop.201600077 2017

  14. [14]

    Population Transfer in a Lambda System In- duced by Detunings

    P. G. Di Stefano et al. “Population Transfer in a Lambda System In- duced by Detunings”. In:Physical Review B91.22 (2015), p. 224506. doi:10.1103/PhysRevB.91.224506

    doi:10.1103/physrevb.91.224506 2015

  15. [15]

    Coherent Trapping in Small Quantum Networks

    T. J. Pope et al. “Coherent Trapping in Small Quantum Networks”. In:Journal of Statistical Mechanics: Theory and Experiment2019.12 (2019), p. 124024.issn: 1742-5468.doi:10.1088/1742-5468/ab54b7

    work page doi:10.1088/1742-5468/ab54b7 2019

  16. [16]

    Atomic Physics and Quantum Optics Using Superconducting Circuits

    J. Q. You and F. Nori. “Atomic Physics and Quantum Optics Using Superconducting Circuits”. In:Nature474 (2011), pp. 589–597.issn: 0028-0836, 1476-4687.doi:10.1038/nature10122

    doi:10.1038/nature10122 2011

  17. [17]

    Optical Selection Rules and Phase-Dependent Adiabatic State Control in a Superconducting Quantum Circuit

    Y. Liu et al. “Optical Selection Rules and Phase-Dependent Adiabatic State Control in a Superconducting Quantum Circuit”. In:Physical Review Letters95 (2005), p. 087001.doi:10.1103/PhysRevLett.95. 087001

    doi:10.1103/physrevlett.95 2005

  18. [18]

    Observation of chiral solitary waves in a nonlinear Aharonov-Bohm ring

    I. Velkovsky et al. “Observation of chiral solitary waves in a nonlinear Aharonov-Bohm ring”. In:arXiv preprint arXiv:2406.01732(2024). 14

    Pith/arXiv arXiv 2024

  19. [19]

    Machine Learning and Quantum Devices

    F. Marquardt. “Machine Learning and Quantum Devices”. In:SciPost Physics Lecture Notes(2021), p. 29.issn: 2590-1990.doi:10.21468/ SciPostPhysLectNotes.29

    2021

  20. [20]

    Artificial intelligence and machine learning for quan- tum technologies

    M. Krenn et al. “Artificial intelligence and machine learning for quan- tum technologies”. In:Phys. Rev. A107 (1 2023), p. 010101.doi: 10.1103/PhysRevA.107.010101

    doi:10.1103/physreva.107.010101 2023

  21. [21]

    Reinforcement learning-enhanced protocols for co- herent population-transfer in three-level quantum systems

    J. Brown et al. “Reinforcement learning-enhanced protocols for co- herent population-transfer in three-level quantum systems”. In:New Journal of Physics23.9 (2021), p. 093035.doi:10.1088/1367-2630/ ac2393

    doi:10.1088/1367-2630/ 2021

  22. [22]

    A Tutorial on Optimal Control and Reinforcement Learning Methods for Quantum Technologies

    L. Giannelli et al. “A Tutorial on Optimal Control and Reinforcement Learning Methods for Quantum Technologies”. In:Physics Letters A 434 (2022), p. 128054.issn: 03759601.doi:10 . 1016 / j . physleta . 2022.128054

    arXiv 2022

  23. [23]

    Noise Classification in Three-Level Quantum Net- works by Machine Learning

    S. Mukherjee et al. “Noise Classification in Three-Level Quantum Net- works by Machine Learning”. In:Machine Learning: Science and Tech- nology5.4 (2024), p. 045049.issn: 2632-2153.doi:10 . 1088 / 2632 - 2153/ad9193

    2024

  24. [24]

    Fasone et al.Detection of Noise Correlations in Two Qubit Systems by Machine Learning

    D. Fasone et al.Detection of Noise Correlations in Two Qubit Systems by Machine Learning. 2025.doi:10.48550/arXiv.2509.03389. arXiv: 2509.03389 [quant-ph]

    Pith/arXiv arXiv doi:10.48550/arxiv.2509.03389 2025

  25. [25]

    Artificial Gauge Fields with Ultracold Atoms

    V. Galitski, G. Juzeli¯ unas, and I. B. Spielman. “Artificial Gauge Fields with Ultracold Atoms”. In:Physics Today72.1 (2019), pp. 38–44.issn: 0031-9228, 1945-0699.doi:10.1063/PT.3.4111

    work page doi:10.1063/pt.3.4111 2019

  26. [26]

    Atomic Interferometers: Phase-dependence in Mul- tilevel Atomic Transitions

    S. J. Buckle et al. “Atomic Interferometers: Phase-dependence in Mul- tilevel Atomic Transitions”. In:Optica Acta: International Journal of Optics33.9 (1986), pp. 1129–1140.issn: 0030-3909.doi:10 . 1080 / 713822082

    1986

  27. [27]

    Laser-Induced Adiabatic Atomic Reorientation with Control of Diabatic Losses

    R. G Unanyan et al. “Laser-Induced Adiabatic Atomic Reorientation with Control of Diabatic Losses”. In:Optics Communications139 (1997), pp. 48–54.issn: 00304018.doi:10.1016/S0030-4018(97)00099-0

    work page doi:10.1016/s0030-4018(97)00099-0 1997

  28. [28]

    Burkov.The Hundred-Page Machine Learning Book

    A. Burkov.The Hundred-Page Machine Learning Book. Vol. 1. 2019

    2019

  29. [29]

    D. P. Kingma and J. Ba.Adam: A Method for Stochastic Optimization. 2014.doi:10.48550/arXiv.1412.6980. arXiv:1412.6980 [cs.LG]

    Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980 2014

  30. [30]

    Abadi et al.TensorFlow: Large-Scale Machine Learning on Het- erogeneous Systems

    M. Abadi et al.TensorFlow: Large-Scale Machine Learning on Het- erogeneous Systems. Software available from tensorflow.org. 2015.url: https://www.tensorflow.org/

    2015

  31. [31]

    Chiral Ground-State Currents of Interacting Pho- tons in a Synthetic Magnetic Field

    P. Roushan et al. “Chiral Ground-State Currents of Interacting Pho- tons in a Synthetic Magnetic Field”. In:Nature Physics13 (Feb. 2017), pp. 146–151.issn: 1745-2473, 1745-2481.doi:10.1038/nphys3930. 15

    doi:10.1038/nphys3930 2017

  32. [32]

    Phase-Controlled Coherent Dynamics of a Single Spin under Closed-Contour Interaction

    A. Barfuss et al. “Phase-Controlled Coherent Dynamics of a Single Spin under Closed-Contour Interaction”. In:Nature Physics14 (Nov. 2018), pp. 1087–1091.issn: 1745-2481.doi:10.1038/s41567-018-0231-8. 16

    work page doi:10.1038/s41567-018-0231-8 2018