Pith. sign in

REVIEW 3 major objections 3 minor 2 cited by

Quantum optical shallow networks

T0 review · 3 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a Hong-Ou-Mandel interferometer evaluates a shallow neural network's hidden layer with a constant number of photons, independent of input size and neuron count.

desk verdict Correct HOM identity and a useful M-neuron extension, but the universality claim is not supported—the model is a PSD quadratic form, not a universal approximator. read the letter →

arxiv 2507.21036 v3 pith:5JS5F6A4 submitted 2025-07-28 quant-ph

Simone Roncallo , Angela Rosy Morgillo , Seth Lloyd , Chiara Macchiavello , Lorenzo Maccone This is my paper · ORCID
classification quant-ph
keywords quantumopticalneuralnetworkHong-Ou-Mandeleffectuniversalapproximationshallowsingle-photoninterferenceimageclassificationphotonresourcescalingmachinelearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper aims to show that a single Hong-Ou-Mandel interferometer, using two single photons, one beam splitter, and two bucket detectors, can evaluate the entire hidden layer of a classical shallow neural network. The output is read from the coincidence rate: for an input state $|I\rangle$ and a mixture of $M$ single-photon states with weights $w_i$, that rate encodes $f_{wW}(I)=\sum_i w_i |\langle I,W_{\lambda_i}\rangle|^2$, which the authors identify with a one-hidden-layer network using square-modulus activation, no biases, and positivity/normalization constraints. Once trained, inference costs $O(\varepsilon^{-2}\log(1/\delta))$ photons independent of input size $N$ and neuron count $M$, which the paper calls a superexponential speedup over the classical $O(MN)$ cost. Numerical simulations on MNIST, Fashion-MNIST, and CIFAR-10 show accuracy that improves with $M$ and matches an unconstrained classical network. The strongest intended consequence is that a fixed optical apparatus can classify images with constant optical resources per query.

What carries the argument

The load-bearing object is Eq. (5), the Hong-Ou-Mandel coincidence identity $f_{wW}(I)=\sum_i w_i |\langle I,W_{\lambda_i}\rangle|^2$. The hidden layer of $M$ neurons is encoded as a density operator: a convex mixture of $M$ single-photon states, each carrying $N$ trainable spectral amplitudes, while the input features are encoded in another single-photon state. The beam splitter turns photon indistinguishability into the nonlinear square-modulus operation, and bucket-detector coincidence counting reads out the weighted sum. This object does the work of collapsing an $O(MN)$ matrix-vector computation into a single optical measurement, with the sample-complexity bound $O(\varepsilon^{-2}\log(2/\delta))$ from Hoeffding's inequality supplying the constant number of photons.

What would settle it

Use the paper's own training setup to fit the indefinite quadratic target $f(x,y)=x^2-y^2$ on the unit square with increasing $M$: if the best test error does not go to zero, the constrained family is not a universal approximator; separately, an experiment comparing measured coincidence rates with Eq. (5) for prepared input and mixture states would test the optical identity itself.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the Hong-Ou-Mandel coincidence probability is not merely an overlap measurement but a neural-network evaluation. Feeding a pure single-photon input and a mixed single-photon hidden state $\rho_U = \sum_i w_i |W_{\lambda_i}\rangle\langle W_{\lambda_i}|$ through a balanced beam splitter gives $p(1_a\cap 1_b)=[1-f_{wW}(I)]/2$, so measuring the coincidence rate is equivalent to computing the hidden layer of a shallow network. The model is Eq. (5): $f_{wW}(I)=\sum_i w_i |\langle I,W_{\lambda_i}\rangle|^2$, with each $|W_{\lambda_i}\rangle$ an $N$-parameter neuron and $w_i$ positive output weights summing to one. The authors treat this as a universal approximator in the Hornik-Stinchcombe-White sense, with square-modulus activation and no biases, subject to $L^2$ normalization of the hidden rows and $L^1$ positivity of the output weights. They additionally show that the alternative coherent superposition encoding collapses to a single quadratic neuron and does not gain accuracy with $M$.

Load-bearing premise

The argument assumes that the constrained family of networks, with positive output weights summing to one, unit-norm hidden rows, square-modulus activation, and no hidden biases, still satisfies the Hornik-Stinchcombe-White universality theorem, a proof of which the paper defers to future work.

Editorial extensions

If this is right

  • A trained network can classify a new image with $O(\varepsilon^{-2}\log(2/\delta))$ photons, independent of how many pixels or neurons the network has.
  • The numerical benchmarks show accuracy increasing with $M$: CIFAR-10 planes-versus-dogs rises from 71.2% at $M=2$ to 78.7% at $M=256$, close to the unconstrained classical network's 74.7-78.4% range.
  • The coherent superposition encoding does not scale with $M$; it is equivalent to a single quadratic neuron, so mixtures are the essential ingredient for expressivity.
  • Photon loss can be absorbed as a renormalization hyperparameter $C=\|I\|^2 \operatorname{Tr}[\rho_U]$, and a bias plus sigmoid in post-processing completes the classifier.
  • Because the mixture can be generated by sampling pure states agnostically, training can be done classically and only the final inference needs the optical apparatus.

Reading between the lines

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

  • A direct test of the universality claim would be to train the constrained family to approximate an indefinite quadratic form such as $x^2-y^2$ on the unit square; if the error has a positive floor as $M$ grows, the constrained square-modulus family is not dense, contradicting the asserted universality.
  • The constant-resource guarantee should carry over to other scalar losses of $f_{wW}(I)$, so the same interferometer could serve as an optical kernel evaluator for kernel methods, not only for binary classification.
  • If the universality gap is real, the practical claim narrows from approximating any continuous function to approximating the learnable positive-semidefinite quadratic family, which may still suffice for the low-photon binary imaging tasks the paper targets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The manuscript proposes a Hong-Ou-Mandel interferometer in which one single-photon state encodes an N-feature input and a mixed single-photon state encodes M hidden neurons with probabilities w_i. The authors derive, in Appendix A, that the coincidence probability yields f_wW(I)=Σ_i w_i |⟨I,W_i⟩|^2, which they identify with a classical shallow network and, invoking Hornik-Stinchcombe-White, claim is a universal approximator with O(ε^{-2} log(2/δ)) photons per inference independent of N and M. They support this with constrained gradient-descent training and numerical classification experiments on MNIST, Fashion-MNIST, and CIFAR-10, and conclude that the protocol gives a superexponential speedup over classical shallow networks.

Significance. The paper has genuine strengths: Appendix A gives a self-contained derivation of the coincidence rate from the beam-splitter unitary, Appendix B provides explicit training gradients, Appendix C lays out a sample-complexity argument, the code is openly available, and the numerical comparison with constrained classical networks is a useful sanity check. If the central claims were correct, a constant-photon classifier whose output is a trainable function of the input would be a significant contribution to quantum optical inference. However, the central universal-approximation claim is not merely unproved but false as stated: the constrained family is finite-dimensional and even, so the Hornik-Stinchcombe-White theorem does not apply. The resource claim also omits the cost of encoding the input and hidden parameters. As it stands, the contribution is a constant-shot estimator of a PSD quadratic form, not a universal approximator.

major comments (3)
  1. [Section II, Eq. (5); Appendix A] The identification of f_wW(I)=Σ_i w_i |⟨I,W_i⟩|^2 with a universal approximator following Hornik-Stinchcombe-White is invalid for the constrained family. Under the SLM discretization used in Section II, f_wW(x)=Σ_i w_i |Σ_j (W_i)_j x_j|^2 = x^T A x with A=Σ_i w_i W_i W_i^T, a positive semidefinite matrix because w_i≥0. Thus every representable pre-activation is a PSD quadratic form, the model output F_θ(x)=σ(f_wW(x)+β) is even under x→−x, and the representable functions lie in a finite-dimensional subset of C(K) (dimension N(N+1)/2), so they cannot be dense. For instance, on K=[-1,1]^N the linear function x_1 is at uniform distance at least 1 from every model output. Hornik-Stinchcombe-White requires hidden biases and unconstrained output weights and therefore does not apply. The Conclusions explicitly defer a rigorous characterization, so the central universal-approximation claim is unsupported.
  2. [Section II, 'constant resources'; Section C] The claimed constant resource cost counts only the number of photons needed to estimate p(1_a∩1_b) and the post-processing operations. It does not include the resources needed to encode the input: for a digital image encoded on an N-pixel SLM (Section II and Section III), writing the N pixel amplitudes requires O(N) resources, and preparing the hidden mixture with M components of N parameters requires O(MN) parameters and operations. Hence the abstract's statement that 'once trained, our model requires constant optical resources regardless of the number of input features and neurons' is not supported by the analysis as stated. If the claim is only that the number of measurement shots is O(ε^{-2} log(2/δ)) independent of N and M, this should be stated explicitly and the comparison with the classical O(MN) inference cost adjusted accordingly.
  3. [Section III, Table I and Fig. 3] The numerical benchmarks show that the constrained model achieves good accuracy on three binary tasks and that accuracy increases with M, but this does not provide evidence for universal approximation. Since the model family is contained in the set of PSD quadratic forms (see the first comment), any dataset that is separable by such forms can show this scaling; the experiments do not address approximation of arbitrary continuous functions or of functions outside the even/PSD class.
minor comments (3)
  1. [Section C, Eqs. (C3)-(C5)] The Hoeffding inequality in Theorem 1 is stated for random variables bounded in [0,1], but it is applied to coincidence outcomes bounded in [0,1/2]. The correct bound is δ ≤ 2 exp(-8 n ε²), not 2 exp(-2 n ε²); the asymptotic O(ε^{-2} log(2/δ)) scaling is unchanged.
  2. [Section II, Eq. (5)] The text says that Eq. (5) is 'precisely the response function of a classical shallow neural network' and that 'we obtain Eq. (1)' for a square absolute value activation with no biases, but Eq. (1) as written includes hidden biases β_i and a general activation g. Since the model has no hidden biases, the claimed identification with Eq. (1) is misleading and should be stated as a constrained specialization.
  3. [Appendix A, Eq. (A11)] The hyperparameter C is introduced as accounting for optical losses, but in the subsequent definitions f(I)=C-2p(1_a∩1_b), C is fully determined by C=∥I∥²Tr[ρ_Uλ]. Please clarify whether C is a fixed physical normalization or a trainable/free parameter, since the text is ambiguous.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the HOM response is re-derived from Wick's theorem; the constrained-family universality gap is an unsupported mathematical claim, not a circular reduction.

full rationale

The paper's central physical derivation is self-contained. In Supplementary A, the coincidence probability is computed from Wick contractions (Eqs. A8–A11), yielding p(1_a∩1_b) = [||I||² Trρ − Σ_i w_i |⟨I,W_i⟩|²]/2, so Eq. (5) follows from the Hong-Ou-Mandel mechanism rather than being a fitted input. The single-neuron response f_W = |⟨I,W⟩|² is re-derived in Eqs. (A12)–(A13), not merely imported from the same authors' earlier [38]. The sample-complexity bound O(ε⁻² log(2/δ)) is derived from Hoeffding's inequality, and no trainable parameter is renamed as a prediction; test accuracy is obtained by ordinary training and evaluation on external benchmarks (MNIST, Fashion-MNIST, CIFAR-10), with classical comparisons in Supplementary D. The self-citations ([38], [39]) are minor and not load-bearing: [38] supports speedup framing whose ingredients are re-derived here, and [39] is an implementation explicitly stated to be under investigation. The main caveat is a correctness issue, not circularity: identifying Eq. (A17) with Eq. (1) and invoking Hornik–Stinchcombe–White [3] is mathematically unsupported, since the square-modulus, bias-free, positive-normalized, L2-normalized family is finite-dimensional and cannot be dense in C(K). The paper itself concedes the gap by listing 'a rigorous mathematical characterization of the universal approximation capabilities' as a future direction. Because the derivation chain does not reduce to its own inputs, the circularity score is low.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The derivation of the HOM response is self-contained, but the central claims rest on two unproven modeling assumptions: HSW universality for a constrained quadratic family, and treating N-mode single-photon state preparation as O(1). The paper's own conclusions list rigorous universal approximation as future work.

free parameters (1)
  • Normalization hyperparameter C = set to ||I||^2 Tr[rho_U]; not fitted in simulations
    Introduced in Eq. (A11) to absorb optical losses; the response f = C - 2p depends on it, but it is a physical normalization rather than a fitted constant. It would need calibration in a real experiment.
assumptions (4)
  • ad hoc to paper Hornik-Stinchcombe-White universality applies to f_wW(I) = sum_i w_i |<I,W_i>|^2 with w_i in the simplex and unit-norm W_i, no hidden biases.
    Invoked in Section II and Appendix A; the actual family is PSD quadratic forms, a finite-dimensional set, so the theorem's hypotheses are not met and no proof is given.
  • ad hoc to paper Preparing the N-mode input single-photon state and the M-component hidden mixture costs O(1) physical resources during inference.
    Section II and Section C count only photons and classical postprocessing; N-pixel SLM encoding, generation of the mixture, and N-mode interference devices are excluded from the resource count.
  • domain assumption The two photons are monochromatic, same-frequency, and the single-photon approximation holds with an ideal balanced beam splitter and bucket detectors.
    Appendix A uses Eq. (A4) and Wick contractions; these are standard quantum optics assumptions, reasonable but untested experimentally in this paper.
  • standard math Hoeffding's inequality applies to the 0/1 coincidence samples from the mixture.
    Section C; standard use of Hoeffding's inequality for bounded random variables.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum optical shallow networks." pith.science (2026). https://pith.science/paper/5JS5F6A4

@misc{pith2026250721036,
  author       = {Pith},
  title        = {Pith review of: Quantum optical shallow networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5JS5F6A4}},
  note         = {Machine review of arXiv:2507.21036}
}
read the original abstract

Classical shallow networks are universal approximators. Given a sufficient number of neurons, they can reproduce any continuous function to arbitrary precision, with a resource cost that scales linearly in both the input size and the number of trainable parameters. In this work, we present a quantum optical protocol that implements a shallow network with an arbitrary number of neurons. Both the input data and the parameters are encoded into single-photon states. Leveraging the Hong-Ou-Mandel effect, the network output is determined by the coincidence rates measured when the photons interfere at a beam splitter, with multiple neurons prepared as a mixture of single-photon states. Remarkably, once trained, our model requires constant optical resources regardless of the number of input features and neurons.

Figures

Figures reproduced from arXiv: 2507.21036 by the authors.

Figure 1
Figure 1. FIG. 1. Quantum optical shallow network implemented by [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Analogy between a classical shallow neural network, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. History plot of the quantum optical shallow network, when classifying images of planes and dogs from the CIFAR-10 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Quantum optical neural networks using atom-cavity interactions to provide all-optical nonlinearity

    quant-ph 2025-11 conditional novelty 6.0 of 10

    A simulated neural network uses atom-cavity two-level neurons as all-optical nonlinear activations and reports ~95% accuracy on MNIST and SAT-6.

  2. Quantum Optical Reinforcement Learning via Spectrum-Resolved Hong-Ou-Mandel Interference

    quant-ph 2026-07 conditional novelty 5.0 of 10

    A spectrum-resolved HOM interference readout, mapped into an actor-critic, is claimed to outperform matching MLP agents on continuous-control benchmarks and to restore drifted transmon-gate fidelities in simulation.

Reference graph

Works this paper leans on

52 extracted references · 29 canonical work pages · cited by 2 Pith papers

  1. [38]

    Roncallo, A

    S. Roncallo, A. R. Morgillo, C. Macchiavello, L. Maccone, and S. Lloyd, Quantum optical classifier with superexpo- nential speedup, Commun. Phys8, 147 (2025)

    2025

  2. [1]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, and A. Courville,Deep Learn- ing(MIT Press, 2016)http://www.deeplearningbook. org

    2016

  3. [2]

    Cybenko, Approximation by superpositions of a sig- moidal function, Math

    G. Cybenko, Approximation by superpositions of a sig- moidal function, Math. Control Signals Syst.2, 303 (1989)

    work page 1989

  4. [3]

    Hornik, M

    K. Hornik, M. Stinchcombe, and H. White, Multilayer feedforward networks are universal approximators, Neu- ral Netw.2, 359 (1989)

    work page 1989

  5. [4]

    Hornik, Approximation capabilities of multilayer feed- forward networks, Neural Netw.4, 251 (1991)

    K. Hornik, Approximation capabilities of multilayer feed- forward networks, Neural Netw.4, 251 (1991)

    1991

  6. [5]

    H. N. Mhaskar and C. A. Micchelli, Approximation by su- perposition of sigmoidal and radial basis functions, Adv. Appl. Math.13, 350 (1992)

    work page 1992

  7. [6]

    Leshno, V

    M. Leshno, V. Y. Lin, A. Pinkus, and S. Schocken, Mul- tilayer feedforward networks with a nonpolynomial ac- tivation function can approximate any function, Neural Netw.6, 861 (1993)

    work page 1993

  8. [7]

    A. R. Barron, Universal approximation bounds for su- perpositions of a sigmoidal function, IEEE Trans. Inf. Theory39, 930 (1993)

    work page 1993

Show all 52 references
  1. [8]

    Chen and H

    T. Chen and H. Chen, Universal approximation to non- linear operators by neural networks with arbitrary activa- tion functions and its application to dynamical systems, IEEE Trans. Neural Netw6, 911 (1995)

    1995

  2. [9]

    Lecun, L

    Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, Gradient-based learning applied to document recogni- tion, Proc. IEEE86, 2278 (1998)

    1998

  3. [10]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, ImageNet classification with deep convolutional neural networks, Commun. ACM60, 84–90 (2017)

    2017

  4. [11]

    K. He, X. Zhang, S. Ren, and J. Sun, Deep residual learn- ing for image recognition, inIEEE Conference on Com- puter Vision and Pattern Recognition, CVPR ’16, p. 770

  5. [12]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, An im- age is worth 16x16 words: Transformers for image recog- nition at scale, inInternational Conference on Learnin...

    2010 arXiv

  6. [13]

    L. Cai, J. Gao, and D. Zhao, A review of the applica- tion of deep learning in medical image classification and segmentation, Ann. Transl. Med.8, 713 (2020)

    2020

  7. [14]

    Dao and N

    L. Dao and N. Q. Ly, Recent advances in medical image classification, Int. J. Adv. Comput. Sci.15, 7 (2024)

    2024

  8. [15]

    R. G. Babu, K. Dhineshkumar, R. Sharma, and R. Kr- ishnamoorthy, A survey of machine learning techniques using for image classification in home security, inIOP Conference Series: Materials Science and Engineering, Vol. 1055 (2021) p. 012088

    2021

  9. [16]

    Z. Long, R. Mccreadie, and M. Imran, Crisisvit: A robust 13 vision transformer for crisis image classification (2024), arXiv:2401.02838

    2024 arXiv

  10. [17]

    Amerini, C.-T

    I. Amerini, C.-T. Li, and R. Caldelli, Social network iden- tification through image classification with CNN, IEEE Access7, 35264 (2019)

    2019

  11. [18]

    Lloyd, M

    S. Lloyd, M. Mohseni, and P. Rebentrost, Quantum algo- rithms for supervised and unsupervised machine learning (2013), arXiv:1307.0411 [quant-ph]

    2013 arXiv

  12. [19]

    X.-D. Cai, D. Wu, Z.-E. Su, M.-C. Chen, X.-L. Wang, L. Li, N.-L. Liu, C.-Y. Lu, and J.-W. Pan, Entanglement- based machine learning on a quantum computer, Phys. Rev. Lett.114, 110504 (2015)

    2015

  13. [20]

    Tacchino, C

    F. Tacchino, C. Macchiavello, D. Gerace, and D. Bajoni, An artificial neuron implemented on an actual quantum processor, Npj Quantum Inf.5, 26 (2019)

    2019

  14. [21]

    Benatti, S

    F. Benatti, S. Mancini, and S. Mangini, Continuous variable quantum perceptron, Int. J. Quantum Inf.17, 1941009 (2019)

    2019

  15. [22]

    Mangini, F

    S. Mangini, F. Tacchino, D. Gerace, C. Macchiavello, and D. Bajoni, Quantum computing model of an artifi- cial neuron with continuously valued input data, Mach. Learn.: Sci. Technol.1, 045008 (2020)

    2020

  16. [23]

    G. R. Steinbrecher, J. P. Olson, D. Englund, and J. Car- olan, Quantum optical neural networks, Npj Quantum Inf.5, 60 (2019)

    2019

  17. [24]

    Killoran, T

    N. Killoran, T. R. Bromley, J. M. Arrazola, M. Schuld, N. Quesada, and S. Lloyd, Continuous-variable quantum neural networks, Phys. Rev. Res.1, 033063 (2019)

    2019

  18. [25]

    X. Sui, Q. Wu, J. Liu, Q. Chen, and G. Gu, A review of optical neural networks, IEEE Access8, 70773 (2020)

    2020

  19. [26]

    Zhang, H

    A. Zhang, H. Zhan, J. Liao, K. Zheng, T. Jiang, M. Mi, P. Yao, and L. Zhang, Quantum verification of NP prob- lems with single photons and linear optics, Light Sci. Appl.10, 169 (2021)

    2021

  20. [27]

    Spall, X

    J. Spall, X. Guo, and A. I. Lvovsky, Hybrid training of optical neural networks, Optica9, 803 (2022)

    2022

  21. [28]

    Stanev, N

    D. Stanev, N. Spagnolo, and F. Sciarrino, Deterministic optimal quantum cloning via a quantum-optical neural network, Phys. Rev. Res.5, 013139 (2023)

    2023

  22. [29]

    C. Wood, S. Shrapnel, and G. J. Milburn, A Kerr ker- nel quantum learning machine (2024), arXiv:2404.01787 [quant-ph]

    2024 arXiv

  23. [30]

    Spall, X

    J. Spall, X. Guo, and A. I. Lvovsky, Training neural networks with end-to-end optical backpropagation, Adv. Photonics7, 016004 (2025)

    2025

  24. [31]

    F. Hoch, E. Caruccio, G. Rodari, T. Francalanci, A. Suprano, T. Giordani, G. Carvacho, N. Spagnolo, S. Koudia, M. Proietti,et al., Quantum machine learn- ing with adaptive boson sampling via post-selection, Nat. Commun.16, 902 (2025)

    2025

  25. [32]

    Slabbert and F

    D. Slabbert and F. Petruccione, Classical-quantum ap- proach to image classification: Autoencoders and quan- tum SVMs, A VS Quantum Sci.7, 023804 (2025)

    2025

  26. [33]

    Y. Sun, D. Li, Q. Xiang, Y. Yuan, Z. Hu, X. Hua, Y. Jiang, Y. Zhu, and Y. Fu, Scalable quantum convo- lutional neural network for image classification, Phys. A: Stat. Mech. Appl.657, 130226 (2025)

    2025

  27. [34]

    Sakurai, A

    A. Sakurai, A. Hayashi, W. J. Munro, and K. Nemoto, Quantum optical reservoir computing powered by boson sampling, Opt. Quantum3, 238 (2025)

    2025

  28. [35]

    Hong, Z.-Y

    C.-K. Hong, Z.-Y. Ou, and L. Mandel, Measurement of subpicosecond time intervals between two photons by in- terference, Phys. Rev. Lett.59, 2044 (1987)

    1987

  29. [36]

    A. M. Bra´ nczyk, Hong-Ou-Mandel interference (2017), arXiv:1711.00080

    2017 arXiv

  30. [37]

    Bowie, S

    C. Bowie, S. Shrapnel, and M. J. Kewming, Quan- tum kernel evaluation via Hong–Ou–Mandel interference, Quantum Sci. Technol.9, 015001 (2023)

    2023

  31. [39]

    Minati, S

    G. Minati, S. Roncallo, S. Scrofana, A. R. Morgillo, N. Spagnolo, C. Macchiavello, L. Maccone, V. Ci- mini, and F. Sciarrino, Quantum optical neuron for im- age classification via multiphoton interference (2026), arXiv:2603.28879

    2026

  32. [40]

    N. Crum, M. M. Hassan, A. Green, and G. Siopsis, Mode distinguishability in multi-photon interference (2025), arXiv:2501.14915

    2025 arXiv

  33. [41]

    Hoeffding, Probability inequalities for sums of bounded random variables, J

    W. Hoeffding, Probability inequalities for sums of bounded random variables, J. Am. Stat. Assoc.58, 13 (1963)

    1963

  34. [42]

    Robbins and S

    H. Robbins and S. Monro, A stochastic approximation method, Ann. Math. Stat.22, 400 (1951)

    1951

  35. [43]

    Salimans and D

    T. Salimans and D. P. Kingma, Weight normalization: A simple reparameterization to accelerate training of deep neural networks, inAdv. Neural Inf. Process., Vol. 29 (2016)

    2016

  36. [44]

    LeCun, C

    Y. LeCun, C. Cortes, and C. J. Burges,https://www. tensorflow.org/datasets/catalog/mnist

  37. [45]

    H. Xiao, K. Rasul, and R. Vollgraf,https://www. tensorflow.org/datasets/catalog/fashion_mnist

  38. [46]

    Krizhevsky,https://www.tensorflow.org/ datasets/catalog/cifar10

    A. Krizhevsky,https://www.tensorflow.org/ datasets/catalog/cifar10

  39. [47]

    D. P. Kingma and J. Ba, Adam: A method for stochastic optimization (2014), arXiv:1412.6980

    2014 arXiv

  40. [48]

    Glorot and Y

    X. Glorot and Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, inProceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, Vol. 9 (2010) pp. 249–256

    2010

  41. [49]

    A. R. Morgillo and S. Roncallo,https://github.com/ simoneroncallo/quantum-optical-network

  42. [50]

    Arunachalam and R

    S. Arunachalam and R. De Wolf, Optimal quantum sam- ple complexity of learning algorithms, J. Mach. Learn. Res.19, 1 (2018)

    2018

  43. [51]

    Cheng, N

    H.-C. Cheng, N. Datta, N. Liu, T. Nuradha, R. Salz- mann, and M. M. Wilde, An invitation to the sample complexity of quantum hypothesis testing, npj Quantum Inf.11, 1 (2025)

    2025

  44. [52]

    Abadi et al., TensorFlow: Large-scale machine learn- ing on heterogeneous systems (2015)

    M. Abadi et al., TensorFlow: Large-scale machine learn- ing on heterogeneous systems (2015)

    2015

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.