Pith. sign in

REVIEW 4 major objections 5 minor 57 references

A continuous-variable photonic quantum extreme learning machine—a fixed Gaussian optical circuit with only a linear readout—classifies top-quark and Higgs events as accurately as small trained neural networks, with nanosecond-scale latency

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

2026-08-04 09:38 UTC pith:WNUM7MEP

load-bearing objection A clean, honest benchmark of a Gaussian photonic ELM on collider data, but the nanosecond-latency claim ignores shot noise in moment estimation and the missing classical baselines leave the photonic contribution unisolated. the 4 major comments →

arxiv 2510.13994 v2 pith:WNUM7MEP submitted 2025-10-15 quant-ph hep-exhep-ph

Continuous-variable photonic quantum extreme learning machines for fast collider-data selection

Benedikt Maier , Michael Spannowsky , Simon Williams This is my paper
classification quant-ph hep-exhep-ph
keywords quantum extreme learning machinecontinuous-variable quantum computingGaussian photonic circuitrandom feature maptop-jet taggingHiggs-boson identificationlinear readoutcollider trigger
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 sets out to show that a continuous-variable photonic quantum extreme learning machine (QELM) can act as a fast, low-overhead classifier for collider data. Event features are encoded as quadrature displacements on optical modes, sent through a fixed Gaussian photonic circuit, and read out as a high-dimensional random feature vector; only a linear classifier is then trained. On top-jet tagging and Higgs-boson identification, this setup outperforms a two-hidden-unit multilayer perceptron at every training size considered and matches or exceeds a ten-hidden-unit perceptron at large sample sizes on the top-jet task, while showing much lower run-to-run variance. The practical payoff, if the claim holds, is a room-temperature, low-power front-end whose retraining is a single regression solve and whose inference latency is set by the optical path and detector response rather than by iterative digital computation.

Core claim

The central claim is that a fixed, randomly drawn Gaussian photonic circuit—a cascade of controlled-addition gates on M qumodes—supplies a random feature map Φ(x) that is expressive enough for collider classification even though no part of the circuit is trained. After displacement encoding of the input features, the output modes are measured by homodyne or photon-number-resolving detection, yielding feature vectors of dimension 5M or 2M, and a single logistic or ridge regression on those features produces the classifier. Against parameter-matched MLP baselines on identical train/validation/test splits, the QELM is reported to beat the small MLP everywhere and to match or beat the larger MLP

What carries the argument

The central object is the continuous-variable quantum extreme learning machine (CV-QELM): a three-stage pipeline of displacement encoding D(α(x)) on M optical modes, propagation through a fixed-time Gaussian 'quantum substrate' Û_G made of cascaded controlled-addition gates with randomly drawn parameters, and Gaussian-compatible readout (homodyne, 5M features, or photon-number-resolving, 2M features) producing the random feature vector. The argument rests on this fixed random feature map plus a linear readout: ridge regression is solved in closed form as (Phi^T Phi + lambda I)^(-1) Phi^T T, so no backpropagation, learning-rate tuning, or early stopping is needed, and the substrate's randomne

Load-bearing premise

The nanosecond-latency and trigger-readiness claims assume that the feature vector—5M quadrature moments for homodyne readout—can be estimated to useful precision within that time, but the paper simulates exact expectation values from covariance matrices and never accounts for the repeated measurement shots needed to suppress finite-shot noise.

What would settle it

Simulate the QELM with finite measurement statistics: for a fixed per-event shot budget (say 10, 100, or 1,000 homodyne measurements), estimate Φ_hom(x) with shot noise and rerun the classification; if the reported accuracy advantage erodes before the shot count that fits in a nanosecond budget, the central speed and trigger claim fails.

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

If this is right

  • On the top-jet tagging benchmark (16 features, up to 10^5 training examples), the QELM outperforms a two-hidden-unit MLP at every training size and surpasses a ten-hidden-unit MLP at the largest sample size, with both readout types and both regression variants.
  • On the Higgs-boson task (10 features), the QELM again beats the two-hidden-unit MLP at all sample sizes; with logistic regression it matches (PNR readout) or slightly exceeds (homodyne readout) the ten-hidden-unit MLP.
  • Training the entire model is a single regularised linear solve, so retraining is fast and deterministic, with no optimisation variance and no hyperparameter iteration.
  • Inference latency is fixed by the optical path and detector response—on the order of nanoseconds for homodyne detection—and is independent of event complexity within the designed optical depth.
  • Run-to-run accuracy variance is substantially lower than the MLP baselines because the feature map is fixed and the readout is fit analytically.

Where Pith is reading between the lines

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

  • Extension: because the Gaussian substrate is classically simulable, the exact same feature map can be reproduced in software; the QELM's practical value therefore reduces to whether the optical path can produce the moments faster and at lower power than a digital simulation—a hardware comparison the paper leaves open.
  • Extension: a testable refinement would add finite-shot statistics—estimating the quadrature means, variances, and covariances to useful precision requires many measurement repetitions per event, and that repetition count, not photon transit time, may set the true per-event latency; degrading the features with shot noise in simulation would settle this directly.
  • Extension: the same recipe—fixed random Gaussian feature map plus linear readout—is portable to other high-throughput tasks in the data-acquisition chain, such as multi-class tagging, regression on calibration quantities, and domain-shift adaptation, since retraining is a single solve.
  • Extension: sweeping substrate depth and squeezing strength would map how the random feature map's expressivity saturates, yielding design scaling laws for a future on-detector implementation.

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

4 major / 5 minor

Summary. The paper introduces a continuous-variable photonic quantum extreme learning machine (CV-QELM) and evaluates it as a fast front-end classifier for collider data. Classical input features are encoded as quadrature displacements of M optical modes; the state evolves through a fixed Gaussian circuit composed of cascaded CX gates; readout is by PNR or homodyne detection; the feature vector consists of exact first- and second-order moments (Eqs. 3.10, 3.13). Only a linear logistic/ridge readout is trained. Using the hls4ml top-tagging dataset (F=16) and the Higgs ML dataset (F=10), the authors report that the QELM consistently outperforms an MLP with H=2 and, at 10^5 training samples, matches or exceeds an H=10 MLP in several configurations. They argue that nanosecond optical latency, analytic retraining, low power and room-temperature operation make the device suitable for first-stage trigger integration.

Significance. If the accuracy comparisons are reproducible, the paper provides a useful benchmark for Gaussian photonic random features in collider data selection: it uses public datasets, identical splits, repeated runs, and explicitly acknowledges classical simulability. The analytic readout and fixed-time substrate are attractive. However, the manuscript's headline claims go beyond the evidence: the features are simulated as noiseless expectation values, no classical random-feature baseline is included, and hyperparameters are unspecified. These gaps affect the central 'trigger-ready, fast inference' claim, though they appear addressable with additional simulations and clarification.

major comments (4)
  1. [Section 3, Eq. (3.13); Section 5] The feature vector in Eq. (3.13) contains ⟨x_i⟩, ⟨p_i⟩, ⟨x_i^2⟩, ⟨p_i^2⟩, and ⟨{x_i,p_i}⟩/2. A homodyne measurement yields one sample of one quadrature per shot, not an expectation value. Estimating these moments to useful precision requires many repeated preparations of the same event, and conjugate quadratures cannot be measured simultaneously on one copy. Since the simulations use exact covariance-matrix expectation values, the reported accuracy is an upper bound for a single-shot/few-shot trigger setting. The claim in Section 5 that 'the optical path and detector response set both the analytic and inference latency' is therefore not supported. The authors should either simulate classifier performance with finite-shot/noisy moment estimates as a function of shots per event, or revise the trigger-readiness claims and position the proposal as offline/post-DAQ processing.
  2. [Section 4, Eqs. (4.1)-(4.2), Figures 3-8] The comparison is not parameter-matched as claimed. For F=16, the homodyne QELM has F+R+1 = 97 trainable parameters versus 37 for the H=2 MLP; for F=10, it has 61 versus 25. The QELM is thus not 'parameter-matched' to H=2. More importantly, a fixed Gaussian circuit is classically simulable and equivalent to a particular random feature map; without comparing to a plain linear model on the raw features or to classical random-feature/random-Fourier features of the same dimension and same training cost, the reported gains cannot be attributed to the photonic substrate. Please add such baselines and re-state parameter comparisons.
  3. [Section 3, around Eq. (3.5); Section 4] The CX gate angles θ_m are 'randomly drawn from experimentally realisable ranges' but the ranges are never specified; the displacement encoding scale α(x) is not defined; the regularization λ is not reported. Accuracy distributions across runs depend on the random substrate ensemble; without distribution definitions or code, results are not reproducible and post-hoc selection of a favorable ensemble cannot be ruled out. Specify these details or release the code/random seeds.
  4. [Abstract vs. Section 4.2, Figures 7-8] The abstract claims the QELM 'matches or exceeds an MLP with ten hidden units at large sample sizes', but for ridge regression on the Higgs task the text states accuracy is 'marginally below' the H=10 MLP for both PNR and homodyne. The abstract should be qualified to the configurations that actually achieve this (logistic readout, and top-tagging task) or state 'comparable'. As written, the claim misrepresents the results.
minor comments (5)
  1. [Section 4.1] Typo: 'high-level observables that encapsulating' should be 'that encapsulate'.
  2. [Section 3, footnote] The footnote has a comma splice ('Note, here we consider') and the homodyne speed statement would benefit from a quantitative reference or bandwidth estimate.
  3. [Section 4] No information is given on preprocessing (standardization, dataset version, class balance) of the hls4ml and Higgs datasets. State these details.
  4. [Figures 4-5, 7-8] The histograms do not specify how 'independent initialisations' are generated (substrate draws, readout initialization, or both). Clarify.
  5. [Section 5] The text repeatedly invokes 'Level-1 trigger constraints' but gives no numerical latency budget against which the nanosecond claim is assessed. Define the target and compare.

Circularity Check

0 steps flagged

No circularity: the CV-QELM results are empirical benchmark accuracies obtained from an explicit feature definition and a linear readout; no prediction reduces to a fitted input or self-citation.

full rationale

The paper's central claims are test-set accuracies on two public datasets (hls4ml and Higgs ML) using identical splits. The feature map is defined constructively in Eqs. (3.5)-(3.13): displacement encoding, fixed Gaussian substrate, and PNR/homodyne moments. The readout is trained by ridge regression (3.16) and evaluated on held-out data; this is a standard train/test evaluation, not a fitted parameter renamed as a prediction. No uniqueness theorem or ansatz is imported from the authors' prior work to force the result; Gaussian operations and QELM limitations are supported by external references, and the paper explicitly acknowledges classical simulability and targets speed rather than quantum advantage. The nanosecond-latency and hardware claims are programmatic—the paper itself defers to future 'on-bench validation of latency, power, and noise characteristics'—and the finite-shot/noise issue is an experimental correctness risk, not a circular reduction. Self-citations appear in the bibliography but are not load-bearing for the benchmark conclusion.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 0 invented entities

The central claim rests on an architecture whose most important knobs (encoding scale, gate-angle ranges, regularization) are undisclosed, and on the assumption of ideal noiseless measurements. The substrate itself introduces no new physical entities.

free parameters (3)
  • CX gate angle range (theta_m)
    The cascade of CX gates in Eq (3.5) uses angles 'randomly drawn from experimentally realisable ranges', but the range/distribution is never specified. This controls the random feature map and hence the classification accuracy.
  • Displacement encoding scale
    Inputs are embedded 'via quadrature displacements' with one feature per qumode (Sec 3, Fig 2), but the paper does not state how feature values x_i are mapped to displacement amplitudes alpha_i (e.g., normalization or scaling). This scaling relative to vacuum noise sets the feature magnitudes.
  • Regularization strength lambda
    lambda appears in Eqs (3.14) and (3.16) but the values used in the experiments are not reported; it is a free hyperparameter for both logistic and ridge readouts.
axioms (3)
  • standard math Gaussian unitary transformations preserve Gaussianity; the output state is fully characterized by first and second moments.
    Used throughout Sec 3 to compute feature vectors via covariance matrices; standard result from Gaussian quantum information [20,21].
  • ad hoc to paper The input state is a product of coherent states (vacuum displaced by alpha(x)), and the measured expectation values correspond to ideal, noiseless, exact moments.
    The paper computes feature vectors as expectation values over the output state without specifying shot noise or detector inefficiency; this is required for the claimed nanosecond latency and low variance.
  • domain assumption The two public datasets (hls4ml LHC jet [43], Higgs ML challenge [55]) with their standard 16/10 engineered features are representative collider benchmarks, and accuracy is the appropriate metric.
    The paper evaluates on these datasets and reports accuracy; appropriateness of accuracy vs ROC/AUC for trigger applications is assumed, not justified.

pith-pipeline@v1.3.0-alltime-deepseek · 15460 in / 18447 out tokens · 162255 ms · 2026-08-04T09:38:33.349600+00:00 · methodology

0 comments
read the original abstract

We study continuous-variable photonic quantum extreme learning machines as fast, low-overhead front-ends for collider data processing. Data is encoded in photonic modes through quadrature displacements and propagated through a fixed-time Gaussian quantum substrate. The final readout occurs through Gaussian-compatible measurements to produce a high-dimensional random feature map. Only a linear classifier is trained, using a single logistic regression, so retraining is fast, and the optical path and detector response set the analytical and inference latency. We evaluate this architecture on two representative classification tasks, top-jet tagging and Higgs-boson identification, with parameter-matched multi-layer perceptron (MLP) baselines. Using standard public datasets and identical train, validation, and test splits, the photonic Quantum Extreme Learning Machine (QELM) outperforms an MLP with two hidden units for all considered training sizes, and matches or exceeds an MLP with ten hidden units at large sample sizes, while training only the linear readout. These results indicate that Gaussian photonic extreme-learning machines can provide compact and expressive random features at fixed latency. The combination of deterministic timing, rapid retraining, low optical power, and room temperature operation makes photonic QELMs a credible building block for online data selection and even first-stage trigger integration at future collider experiments.

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

57 extracted references · 4 canonical work pages

  1. [1]

    Reconstructing charged particle track segments with a quantum-enhanced support vector machine,

    P. Duckett, G. Facini, M. Jastrzebski, S. Malik, T. Scanlon, and S. Rettie, “Reconstructing charged particle track segments with a quantum-enhanced support vector machine,” Phys. Rev. D109(Mar, 2024) 052002.https://link.aps.org/doi/10.1103/PhysRevD.109.052002

    work page doi:10.1103/physrevd.109.052002 2024

  2. [2]

    Quantum pathways for charged track finding in high-energy collisions,

    C. Brown, M. Spannowsky, A. Tapper, S. Williams, and I. Xiotidis, “Quantum pathways for charged track finding in high-energy collisions,” Front. Artif. Intell.7(2024) 1339785, [arXiv:2311.00766 [hep-ph]]

    Pith/arXiv arXiv 2024

  3. [3]

    Anomaly detection in high-energy physics using a quantum autoencoder,

    V. S. Ngairangbam, M. Spannowsky, and M. Takeuchi, “Anomaly detection in high-energy physics using a quantum autoencoder,” Phys. Rev. D105(May, 2022) 095004. https://link.aps.org/doi/10.1103/PhysRevD.105.095004

    doi:10.1103/physrevd.105.095004 2022

  4. [4]

    One particle - one qubit: Particle physics data encoding for quantum machine learning,

    A. Bal, M. Klute, B. Maier, M. Oughton, E. Pezone, and M. Spannowsky, “One particle - one qubit: Particle physics data encoding for quantum machine learning,” Phys. Rev. D112(Oct,

  5. [5]

    Towards a quantum computing algorithm for helicity amplitudes and parton showers,

    K. Bepari, S. Malik, M. Spannowsky, and S. Williams, “Towards a quantum computing algorithm for helicity amplitudes and parton showers,” Phys. Rev. D103no. 7, (2021) 076020, [arXiv:2010.00046 [hep-ph]]

    Pith/arXiv arXiv 2021

  6. [6]

    Quantum walk approach to simulating parton showers,

    K. Bepari, S. Malik, M. Spannowsky, and S. Williams, “Quantum walk approach to simulating parton showers,” Phys. Rev. D106no. 5, (2022) 056002, [arXiv:2109.13975 [hep-ph]]

    Pith/arXiv arXiv 2022

  7. [7]

    Collider events on a quantum computer,

    G. Gustafson, S. Prestel, M. Spannowsky, and S. Williams, “Collider events on a quantum computer,” JHEP11(2022) 035, [arXiv:2207.10694 [hep-ph]]

    Pith/arXiv arXiv 2022

  8. [8]

    Quantum algorithm for Feynman loop integrals,

    S. Ram ´ ırez-Uribe, A. E. Renter ´ ıa-Olivo, G. Rodrigo, G. F. R. Sborlini, and L. Vale Silva, “Quantum algorithm for Feynman loop integrals,” JHEP05(2022) 100, [arXiv:2105.08703 [hep-ph]]

    Pith/arXiv arXiv 2022

  9. [9]

    Quantum simulation of scattering amplitudes and interferences in perturbative QCD,

    H. A. Chawdhry, M. Pellen, and S. Williams, “Quantum simulation of scattering amplitudes and interferences in perturbative QCD,” [arXiv:2507.07194 [hep-ph]]

    Pith/arXiv arXiv

  10. [10]

    A quantum algorithm for the n-gluon MHV scattering amplitude,

    E. Bashore, S. Moretti, and T. Vitos, “A quantum algorithm for the n-gluon MHV scattering amplitude,” [arXiv:2507.14252 [hep-ph]]

    Pith/arXiv arXiv

  11. [11]

    Quantum Algorithms for Quantum Field Theories,

    S. P. Jordan, K. S. M. Lee, and J. Preskill, “Quantum Algorithms for Quantum Field Theories,” Science336(2012) 1130–1133, [arXiv:1111.3633 [quant-ph]]. – 17 –

    Pith/arXiv arXiv 2012

  12. [12]

    Simulating Lattice Gauge Theories within Quantum Technologies,

    M. C. Ba˜ nulset al., “Simulating Lattice Gauge Theories within Quantum Technologies,” Eur. Phys. J. D74no. 8, (2020) 165, [arXiv:1911.00003 [quant-ph]]

    Pith/arXiv arXiv 2020

  13. [13]

    Quantum simulation for high-energy physics,

    C. W. Bauer, Z. Davoudi, A. B. Balantekin, T. Bhattacharya, M. Carena, W. A. de Jong, P. Draper, A. El-Khadra, N. Gemelke, M. Hanada, D. Kharzeev, H. Lamm, Y.-Y. Li, J. Liu, M. Lukin, Y. Meurice, C. Monroe, B. Nachman, G. Pagano, J. Preskill, E. Rinaldi, A. Roggero, D. I. Santiago, M. J. Savage, I. Siddiqi, G. Siopsis, D. Van Zanten, N. Wiebe, Y. Yamauchi...

    2023

  14. [14]

    Enhancing quantum field theory simulations on NISQ devices with Hamiltonian truncation,

    J. Ingoldby, M. Spannowsky, T. Sypchenko, and S. Williams, “Enhancing quantum field theory simulations on NISQ devices with Hamiltonian truncation,” Phys. Rev. D110no. 9, (2024) 096016, [arXiv:2407.19022 [quant-ph]]

    Pith/arXiv arXiv 2024

  15. [15]

    Quantum computing for high-energy physics: State of the art and challenges,

    A. Di Meglio, K. Jansen, I. Tavernelli, C. Alexandrou, S. Arunachalam, C. W. Bauer, K. Borras, S. Carrazza, A. Crippa, V. Croft, R. de Putter, A. Delgado, V. Dunjko, D. J. Egger, E. Fern´ andez-Combarro, E. Fuchs, L. Funcke, D. Gonz´ alez-Cuadra, M. Grossi, J. C. Halimeh, Z. Holmes, S. K¨ uhn, D. Lacroix, R. Lewis, D. Lucchesi, M. L. Martinez, F. Meloni, ...

  16. [16]

    Simulating quantum field theories on near-term quantum devices,

    S. Williams, “Simulating quantum field theories on near-term quantum devices,” Int. J. Mod. Phys. A40no. 26, (2025) 2530010

    2025

  17. [17]

    Real-Time Scattering on Quantum Computers via Hamiltonian Truncation,

    J. Ingoldby, M. Spannowsky, T. Sypchenko, S. Williams, and M. Wingate, “Real-Time Scattering on Quantum Computers via Hamiltonian Truncation,” [arXiv:2505.03878 [quant-ph]]

    Pith/arXiv arXiv

  18. [18]

    Quantum simulation of out-of-equilibrium dynamics in gauge theories,

    J. C. Halimeh, N. Mueller, J. Knolle, Z. Papi´ c, and Z. Davoudi, “Quantum simulation of out-of-equilibrium dynamics in gauge theories,” [arXiv:2509.03586 [quant-ph]]

    Pith/arXiv arXiv

  19. [19]

    Quantum computation over continuous variables,

    S. Lloyd and S. L. Braunstein, “Quantum computation over continuous variables,” Phys. Rev. Lett.82(Feb, 1999) 1784–1787.https://link.aps.org/doi/10.1103/PhysRevLett.82.1784

    doi:10.1103/physrevlett.82.1784 1999

  20. [20]

    Quantum information with continuous variables,

    S. L. Braunstein and P. van Loock, “Quantum information with continuous variables,” Rev. Mod. Phys.77(Jun, 2005) 513–577. https://link.aps.org/doi/10.1103/RevModPhys.77.513

    doi:10.1103/revmodphys.77.513 2005

  21. [21]

    Gaussian quantum information,

    C. Weedbrook, S. Pirandola, R. Garc ´ ıa-Patr´ on, N. J. Cerf, T. C. Ralph, J. H. Shapiro, and S. Lloyd, “Gaussian quantum information,” Rev. Mod. Phys.84no. 2, (2012) 621–669, [arXiv:1110.3234 [quant-ph]]

    Pith/arXiv arXiv 2012

  22. [22]

    Continuous variable quantum information: Gaussian states and beyond,

    G. Adesso, S. Ragy, and A. R. Lee, “Continuous variable quantum information: Gaussian states and beyond,” Open Systems & Information Dynamics21no. 01n02, (2014) 1440001, [arXiv:1401.4679]

    Pith/arXiv arXiv 2014

  23. [23]

    Universal quantum computation with continuous-variable cluster states,

    N. C. Menicucci, P. van Loock, M. Gu, C. Weedbrook, T. C. Ralph, and M. A. Nielsen, “Universal quantum computation with continuous-variable cluster states,” Phys. Rev. Lett.97 no. 11, (2006) 110501, [arXiv:quant-ph/0605198]. – 18 –

    Pith/arXiv arXiv 2006

  24. [24]

    Unsupervised event classification with graphs on classical and photonic quantum computers,

    A. Blance and M. Spannowsky, “Unsupervised event classification with graphs on classical and photonic quantum computers,” JHEP21(2020) 170, [arXiv:2103.03897 [hep-ph]]

    Pith/arXiv arXiv 2020

  25. [25]

    Simulating quantum field theories on continuous-variable quantum computers,

    S. Abel, M. Spannowsky, and S. Williams, “Simulating quantum field theories on continuous-variable quantum computers,” Phys. Rev. A110no. 1, (2024) 012607, [arXiv:2403.10619 [quant-ph]]

    Pith/arXiv arXiv 2024

  26. [26]

    Real-time scattering processes with continuous-variable quantum computers,

    S. Abel, M. Spannowsky, and S. Williams, “Real-time scattering processes with continuous-variable quantum computers,” Phys. Rev. A112no. 1, (2025) 012614, [arXiv:2502.01767 [quant-ph]]

    Pith/arXiv arXiv 2025

  27. [27]

    Qumode Tensor Networks for False Vacuum Decay in Quantum Field Theory,

    S. Abel, M. Spannowsky, and S. Williams, “Qumode Tensor Networks for False Vacuum Decay in Quantum Field Theory,” [arXiv:2506.17388 [quant-ph]]

    arXiv

  28. [28]

    Extreme learning machine: Theory and applications,

    G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew, “Extreme learning machine: Theory and applications,” Neurocomputing70no. 1-3, (2006) 489–501

    2006

  29. [29]

    Extreme learning machine for regression and multiclass classification,

    G.-B. Huang, H. Zhou, X. Ding, and R. Zhang, “Extreme learning machine for regression and multiclass classification,” IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics)42no. 2, (2012) 513–529

    2012

  30. [30]

    A comprehensive theoretical framework for the random feature mapping in extreme learning machines,

    T. Guo, G.-B. Huang, and Y. Chen, “A comprehensive theoretical framework for the random feature mapping in extreme learning machines,” Neurocomputing168(2015) 35–48

    2015

  31. [31]

    Deep learning with coherent nanophotonic circuits,

    Y. Shen, N. C. Harris, S. Skirlo, M. Prabhu, T. Baehr-Jones, M. Hochberg, X. Sun, S. Zhao, H. Larochelle, D. Englund, and M. Soljacic, “Deep learning with coherent nanophotonic circuits,” Nature Photonics11(2017) 441–446, [arXiv:1610.02365]

    Pith/arXiv arXiv 2017

  32. [32]

    Quantum reservoir computing with continuous variables,

    M. Henderson, V. Srikumar, and S. Kais, “Quantum reservoir computing with continuous variables,” Phys. Rev. Lett.125no. 26, (2020) 260503, [arXiv:2008.06653]

    Pith/arXiv arXiv 2020

  33. [33]

    Experimental demonstration of reservoir computing on a silicon photonics chip,

    K. Vandoorne, P. Mechet, T. Van Vaerenbergh, M. Fiers, G. Morthier, D. Verstraeten, B. Schrauwen, J. Dambre, and P. Bienstman, “Experimental demonstration of reservoir computing on a silicon photonics chip,” Nature Communications5(2014) 3541

    2014

  34. [34]

    Krylov expressivity in quantum reservoir computing and quantum extreme learning,

    S. ˇCindrak, L. Jaurigue, and K. L¨ udge, “Krylov expressivity in quantum reservoir computing and quantum extreme learning,” [arXiv:2409.12079]

    arXiv

  35. [35]

    On fundamental aspects of quantum extreme learning machines,

    W. Xiong, G. Facelli, M. Sahebi, O. Agnel, T. Chotibut, S. Thanasilp, and Z. Holmes, “On fundamental aspects of quantum extreme learning machines,” Quantum Machine Intelligence7 no. 1, (2025) 20, [arXiv:2312.15124]

    Pith/arXiv arXiv 2025

  36. [36]

    Quantum Machine Learning in Feature Hilbert Spaces,

    M. Schuld and N. Killoran, “Quantum Machine Learning in Feature Hilbert Spaces,” Phys. Rev. Lett.122(Feb, 2019) 040504. https://link.aps.org/doi/10.1103/PhysRevLett.122.040504

    doi:10.1103/physrevlett.122.040504 2019

  37. [37]

    Supervised learning with quantum-enhanced feature spaces,

    V. Havlicek, A. D. C´ orcoles, K. Temme, A. W. Harrow, A. Kandala, J. M. Chow, and J. M. Gambetta, “Supervised learning with quantum-enhanced feature spaces,” Nature567(2019) 209–212, [arXiv:1804.11326 [quant-ph]]

    Pith/arXiv arXiv 2019

  38. [38]

    Opportunities in quantum reservoir computing and extreme learning machines,

    P. Mujal, R. Mart ´ ınez-Pe˜ na, J. Nokkala, J. Garc ´ ıa-Beni, G. L. Giorgi, M. C. Soriano, and R. Zambrini, “Opportunities in quantum reservoir computing and extreme learning machines,” Advanced Quantum Technologies4no. 8, (2021) 2100027, [arXiv:2102.11831]

    Pith/arXiv arXiv 2021

  39. [39]

    Potential and limitations of quantum extreme learning machines,

    L. Innocenti, S. Lorenzo, I. Palmisano, A. Ferraro, M. Paternostro, and G. M. Palma, – 19 – “Potential and limitations of quantum extreme learning machines,” Communications Physics6 no. 1, (2023) 118, [arXiv:2210.00780]

    Pith/arXiv arXiv 2023

  40. [40]

    State estimation with quantum extreme learning machines beyond the scrambling time,

    M. Vetrano, G. Lo Monaco, L. Innocenti, S. Lorenzo, and G. M. Palma, “State estimation with quantum extreme learning machines beyond the scrambling time,” npj Quantum Information 11no. 1, (2025) 20, [arXiv:2409.06782]

    Pith/arXiv arXiv 2025

  41. [41]

    Quantum extreme learning machines for photonic entanglement witnessing,

    D. Zia, L. Innocenti, G. Minati, S. Lorenzo, A. Suprano, R. Di Bartolo, N. Spagnolo, T. Giordani, V. Cimini, G. M. Palma, A. Ferraro, F. Sciarrino, and M. Paternostro, “Quantum extreme learning machines for photonic entanglement witnessing,” [arXiv:2502.18361 [quant-ph]]

    arXiv

  42. [42]

    Entanglement estimation of werner states with a quantum extreme learning machine,

    H. Assil, A. El Allati, and G. L. Giorgi, “Entanglement estimation of werner states with a quantum extreme learning machine,” Phys. Rev. A111(Feb, 2025) 022412. https://link.aps.org/doi/10.1103/PhysRevA.111.022412

    doi:10.1103/physreva.111.022412 2025

  43. [43]

    HLS4ML LHC Jet dataset (150 particles),

    M. Pierini, J. M. Duarte, N. Tran, and M. Freytsis, “HLS4ML LHC Jet dataset (150 particles),” Jan., 2020.https://doi.org/10.5281/zenodo.3602260

    doi:10.5281/zenodo.3602260 2020

  44. [44]

    Fast inference of deep neural networks in FPGAs for particle physics,

    J. Duarte, S. Han, P. Harris, S. Jindariani, E. Kreinar, B. Kreis, J. Ngadiuba, M. Pierini, R. Rivera, N. Tran, and Z. Wu, “Fast inference of deep neural networks in FPGAs for particle physics,” JINST13no. 07, (2018) P07027, [arXiv:1804.06913]

    Pith/arXiv arXiv 2018

  45. [45]

    Fast convolutional neural networks on FPGAs with hls4ml,

    T. Aarrestadet al., “Fast convolutional neural networks on FPGAs with hls4ml,” Machine Learning: Science and Technology2no. 4, (2021) 045015, [arXiv:2101.05108]

    Pith/arXiv arXiv 2021

  46. [46]

    Randomness in neural networks: An overview,

    S. Scardapane and D. Wang, “Randomness in neural networks: An overview,” Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery7no. 2, (2017) e1200

    2017

  47. [47]

    Random features for large-scale kernel machines,

    A. Rahimi and B. Recht, “Random features for large-scale kernel machines,” inAdvances in Neural Information Processing Systems, vol. 20, pp. 1177–1184. 2008. [arXiv:0709.4483 [cs.LG]]

    Pith/arXiv arXiv 2008

  48. [48]

    Production of photonic universal quantum gates enhanced by machine learning,

    K. K. Sabapathy, H. Qi, J. Izaac, and C. Weedbrook, “Production of photonic universal quantum gates enhanced by machine learning,” Phys. Rev. A100(Jul, 2019) 012326. https://link.aps.org/doi/10.1103/PhysRevA.100.012326

    work page doi:10.1103/physreva.100.012326 2019

  49. [49]

    Conversion of gaussian states to non-gaussian states using photon-number-resolving detectors,

    D. Su, C. R. Myers, and K. K. Sabapathy, “Conversion of gaussian states to non-gaussian states using photon-number-resolving detectors,” Phys. Rev. A100(Nov, 2019) 052301. https://link.aps.org/doi/10.1103/PhysRevA.100.052301

    doi:10.1103/physreva.100.052301 2019

  50. [50]

    Homodyne statistics,

    S. L. Braunstein, “Homodyne statistics,” Phys. Rev. A42(Jul, 1990) 474–481. https://link.aps.org/doi/10.1103/PhysRevA.42.474

    work page doi:10.1103/physreva.42.474 1990

  51. [51]

    Fundamental limits to observations of squeezing via balanced homodyne detection,

    S. L. Braunstein and D. D. Crouch, “Fundamental limits to observations of squeezing via balanced homodyne detection,” Phys. Rev. A43(Jan, 1991) 330–337. https://link.aps.org/doi/10.1103/PhysRevA.43.330

    work page doi:10.1103/physreva.43.330 1991

  52. [52]

    Continuous-variable quantum key distribution with 1 mbps secure key rate,

    D. Huang, D. Lin, C. Wang, W. Liu, S. Fang, J. Peng, P. Huang, and G. Zeng, “Continuous-variable quantum key distribution with 1 mbps secure key rate,” Opt. Express23 no. 13, (Jun, 2015) 17511–17519. https://opg.optica.org/oe/abstract.cfm?URI=oe-23-13-17511

    2015

  53. [53]

    Single-photon detection for long-range imaging and sensing,

    R. H. Hadfield, J. Leach, F. Fleming, D. J. Paul, C. H. Tan, J. S. Ng, R. K. Henderson, and G. S. Buller, “Single-photon detection for long-range imaging and sensing,” Optica10no. 9, – 20 – (Sep, 2023) 1124–1141. https://opg.optica.org/optica/abstract.cfm?URI=optica-10-9-1124

    2023

  54. [54]

    Continuous-variable quantum neural networks,

    N. Killoran, T. R. Bromley, J. M. Arrazola, M. Schuld, N. Quesada, and S. Lloyd, “Continuous-variable quantum neural networks,” Phys. Rev. Research1no. 3, (2019) 033063, [arXiv:1906.07682]

    Pith/arXiv arXiv 2019

  55. [55]

    The Higgs boson machine learning challenge,

    C. Adam-Bourdarios, G. Cowan, C. Germain, I. Guyon, B. K´ egl, and D. Rousseau, “The Higgs boson machine learning challenge,” inProceedings of the NIPS 2014 Workshop on High-energy Physics and Machine Learning, G. Cowan, C. Germain, I. Guyon, B. K´ egl, and D. Rousseau, eds., vol. 42 ofProceedings of Machine Learning Research, pp. 19–55. PMLR, Montreal, C...

    2014

  56. [2024]

    037001.https://link.aps.org/doi/10.1103/PRXQuantum.5.037001

    doi:10.1103/prxquantum.5.037001

  57. [2025]

    076004.https://link.aps.org/doi/10.1103/l8y2-87vq

    doi:10.1103/l8y2-87vq