pith. sign in

arxiv: 1907.05103 · v1 · pith:LV6X2JWWnew · submitted 2019-07-11 · 🪐 quant-ph · cs.LG

Machine Learning Kernel Method from a Quantum Generative Model

Przemys{\l}aw Sadowski This is my paper

Pith reviewed 2026-05-24 23:23 UTC · model grok-4.3

classification 🪐 quant-ph cs.LG
keywords quantum machine learningquantum samplingrandom feature mapsNISQ classifierkernel methodrandom quantum circuits
0
0 comments X

The pith

Sampling from random quantum circuits with parametrized rotations yields feature maps for a competitive classifier.

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

The paper proposes a classifier that draws randomized feature maps by sampling from families of random quantum circuits. These circuits incorporate parametrized rotations whose distribution is chosen to produce the maps. The resulting method is presented as simple to implement with intuitive hyperparameters and is shown to match the accuracy of leading classical random-feature classifiers on standard tasks. A sympathetic reader would care because the approach isolates quantum sampling as the operative step, a task viewed as more feasible for near-term devices than full quantum computation.

Core claim

A quantum sampling based classifier is obtained by using randomized feature maps generated from random quantum circuits with parametrized rotations distribution. This yields a method that performs at least as well as top out-of-the-box classical methods, with quantum sampling as the crucial component.

What carries the argument

Randomized feature map produced by sampling from random quantum circuits containing parametrized rotations.

If this is right

  • The classifier requires only sampling from the proposed circuit family and standard classical post-processing.
  • Performance remains competitive without needing deep or fault-tolerant quantum circuits.
  • Hyperparameters consist of circuit depth, rotation distribution, and number of samples, all directly controllable.
  • Quantum sampling replaces the classical random projection step in kernel approximation.

Where Pith is reading between the lines

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

  • If sampling cost scales favorably with system size, the same construction could be applied to larger feature spaces than classical methods can afford.
  • The approach suggests testing whether other generative quantum models produce feature maps with qualitatively different statistical properties than classical random features.
  • Direct comparison of sample efficiency between the quantum circuits and classical alternatives on identical hardware budgets would clarify the practical threshold for advantage.

Load-bearing premise

The statistical properties of the samples drawn from these circuits produce feature maps whose classification accuracy cannot be matched by classical random feature methods at comparable cost.

What would settle it

A classical random-feature implementation that achieves equal or higher accuracy on the same benchmarks while using no more classical compute time than the quantum sampling step requires.

Figures

Figures reproduced from arXiv: 1907.05103 by Przemys{\l}aw Sadowski.

Figure 1
Figure 1. Figure 1: Original random features generation scheme. The crucial part that [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Quantum part of the scheme in the most general picture. We link [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Random features calculation with quantum circuits. For an input [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Circuit Ans¨atze example. The circuit always begins with [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Score histograms for number of random vectors [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Score histograms for number of random vectors [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Scores obtained for number of random vectors [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
read the original abstract

Recently the use of Noisy Intermediate Scale Quantum (NISQ) devices for machine learning tasks has been proposed. The propositions often perform poorly due to various restrictions. However, the quantum devices should perform well in sampling tasks. Thus, we recall theory of sampling-based approach to machine learning and propose a quantum sampling based classifier. Namely, we use randomized feature map approach. We propose a method of quantum sampling based on random quantum circuits with parametrized rotations distribution. We obtain simple to use method with intuitive hyper-parameters that performs at least equally well as top out-of-the-box classical methods. In short we obtain a competitive quantum classifier with crucial component being quantum sampling -- a promising task for quantum supremacy.

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

3 major / 2 minor

Summary. The manuscript proposes a sampling-based kernel classifier that constructs randomized feature maps by sampling from random quantum circuits whose gates include parametrized rotations. It asserts that the resulting quantum feature maps yield classification performance at least equal to leading classical out-of-the-box methods and that the quantum sampling step is the crucial ingredient for potential quantum supremacy.

Significance. If the performance claim were substantiated by controlled experiments, the work would illustrate a concrete route by which NISQ sampling can be embedded inside a standard random-feature kernel pipeline. The emphasis on a small number of intuitive hyperparameters is a practical strength.

major comments (3)
  1. [Abstract] Abstract: the claim that the method 'performs at least equally well as top out-of-the-box classical methods' is stated without any accuracy numbers, datasets, error bars, or explicit comparison to a classical random-feature baseline (e.g., random Fourier features) using an equivalent number of features or sampling budget.
  2. [Method] Method section (description of the parametrized-rotation distribution): no derivation, concentration bound, or statistical test is supplied showing that the feature statistics induced by the quantum circuits differ in a way that cannot be matched by classical Monte-Carlo sampling of an equivalent random-feature map at the same computational cost.
  3. [Results] Results section: the manuscript supplies neither circuit specifications (qubit count, depth, gate set), nor any tabulated classification accuracies, nor any ablation that isolates the contribution of the quantum sampler versus a purely classical random-feature implementation.
minor comments (2)
  1. [Abstract] The abstract would be clearer if it named the concrete datasets or tasks on which competitiveness is claimed.
  2. [Method] Notation for the distribution over parametrized rotations should be introduced with an explicit equation rather than prose only.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the detailed and constructive report. The comments correctly identify several areas where additional quantitative details and clarifications would strengthen the presentation. We will perform a major revision to incorporate the requested information from our experiments. Point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the method 'performs at least equally well as top out-of-the-box classical methods' is stated without any accuracy numbers, datasets, error bars, or explicit comparison to a classical random-feature baseline (e.g., random Fourier features) using an equivalent number of features or sampling budget.

    Authors: We agree that the abstract would be improved by including concrete metrics. The manuscript reports experiments on standard classification datasets showing performance competitive with classical kernel methods. In revision we will update the abstract to state the specific datasets, accuracy values with error bars, and the comparison against random Fourier features using an equivalent number of features and sampling budget. revision: yes

  2. Referee: [Method] Method section (description of the parametrized-rotation distribution): no derivation, concentration bound, or statistical test is supplied showing that the feature statistics induced by the quantum circuits differ in a way that cannot be matched by classical Monte-Carlo sampling of an equivalent random-feature map at the same computational cost.

    Authors: The parametrized-rotation distribution is chosen because it is efficiently sampled on quantum hardware. The current text does not contain a formal derivation or concentration inequality. We will expand the method section with a clearer description of the distribution together with any available empirical statistical comparisons to classical sampling; a rigorous proof that the statistics cannot be matched classically lies beyond the present empirical scope and will be noted as such. revision: partial

  3. Referee: [Results] Results section: the manuscript supplies neither circuit specifications (qubit count, depth, gate set), nor any tabulated classification accuracies, nor any ablation that isolates the contribution of the quantum sampler versus a purely classical random-feature implementation.

    Authors: We will revise the results section to present the circuit specifications used (qubit count, depth, gate set), tabulated accuracies with error bars, and an explicit ablation that compares the quantum sampler against a classical random-feature implementation under identical feature count and sampling budget. revision: yes

standing simulated objections not resolved
  • A formal concentration bound or statistical test proving that the quantum-induced feature statistics cannot be matched by classical Monte-Carlo sampling at equivalent computational cost

Circularity Check

0 steps flagged

No circularity: derivation self-contained with no self-referential reductions

full rationale

The manuscript proposes a randomized feature map classifier based on sampling from random quantum circuits with parametrized rotations. No equations, fitted parameters, or performance claims in the abstract or described content reduce by construction to prior self-results, self-citations, or input data relabeled as outputs. The central assertion of competitive accuracy is presented as an empirical outcome rather than a mathematical identity or fitted prediction. No load-bearing uniqueness theorems or ansatzes imported via self-citation are invoked. The derivation chain therefore remains independent of its own inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on the domain assumption that quantum sampling is a promising route to advantage and on the modeling choice that random circuits with a tunable rotation distribution will generate useful features; no free parameters or invented entities are explicitly introduced in the abstract.

free parameters (1)
  • parametrized rotations distribution
    The distribution from which rotation angles are drawn is presented as a tunable hyper-parameter whose concrete form is not specified in the abstract.
axioms (1)
  • domain assumption Quantum devices should perform well in sampling tasks
    Stated explicitly as the justification for building a sampling-based rather than gate-based classifier.

pith-pipeline@v0.9.0 · 5638 in / 1267 out tokens · 24779 ms · 2026-05-24T23:23:33.843962+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

18 extracted references · 18 canonical work pages · 2 internal anchors

  1. [1]

    Quantum machine learning

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Rebentrost, Nathan Wiebe, and Seth Lloyd. Quantum machine learning. Nature, 549(7671):195, 2017

    work page 2017

  2. [2]

    Quantum machine learning: a classical perspective

    Carlo Ciliberto, Mark Herbster, Alessandro Davide Ialongo, Massim- iliano Pontil, Andrea Rocchetto, Simone Severini, and Leonard Woss- nig. Quantum machine learning: a classical perspective. Proceedings of 12 Machine Learning Kernel Method from a Quantum Generative Model the Royal Society A: Mathematical, Physical and Engineering Sciences , 474(2209):2017...

    work page 2018

  3. [3]

    Quantum computing in the nisq era and beyond

    John Preskill. Quantum computing in the nisq era and beyond. Quan- tum, 2:79, 2018

    work page 2018

  4. [4]

    Complexity-Theoretic Foundations of Quantum Supremacy Experiments

    Scott Aaronson and Lijie Chen. Complexity-theoretic foundations of quantum supremacy experiments. arXiv preprint arXiv:1612.05903 , 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  5. [5]

    Characterizing quantum supremacy in near-term de- vices

    Sergio Boixo, Sergei V Isakov, Vadim N Smelyanskiy, Ryan Babbush, Nan Ding, Zhang Jiang, Michael J Bremner, John M Martinis, and Hartmut Neven. Characterizing quantum supremacy in near-term de- vices. Nature Physics, 14(6):595, 2018

    work page 2018

  6. [6]

    Variational quantum generators: Generative adversarial quantum machine learning for continuous distributions

    Jonathan Romero and Alan Aspuru-Guzik. Variational quantum gener- ators: Generative adversarial quantum machine learning for continuous distributions. arXiv preprint arXiv:1901.00848 , 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1901

  7. [7]

    Random features for large-scale kernel machines

    Ali Rahimi and Benjamin Recht. Random features for large-scale kernel machines. In Advances in neural information processing systems, pages 1177–1184, 2008

    work page 2008

  8. [8]

    Weighted sums of random kitchen sinks: Replacing minimization with randomization in learning

    Ali Rahimi and Benjamin Recht. Weighted sums of random kitchen sinks: Replacing minimization with randomization in learning. In Advances in neural information processing systems , pages 1313–1320, 2009

    work page 2009

  9. [9]

    Qudit quantum-state tomography

    RT Thew, Kae Nemoto, Andrew G White, and William J Munro. Qudit quantum-state tomography. Physical Review A, 66(1):012303, 2002

    work page 2002

  10. [10]

    Quantum Computation and Quantum Information

    Michael A Nielsen and Isaac L Chuang. Quantum Computation and Quantum Information. Cambridge University Press, 2010

    work page 2010

  11. [11]

    Exploratory analysis of mnist handwritten digit for machine learning modelling

    Mohd Razif Shamsuddin, Shuzlina Abdul-Rahman, and Azlinah Mo- hamed. Exploratory analysis of mnist handwritten digit for machine learning modelling. In International Conference on Soft Computing in Data Science, pages 134–145. Springer, 2018

    work page 2018

  12. [12]

    Quantum kitchen sinks: An algorithm for machine learning on near-term quantum computers

    Christopher Wilson, Johannes Otterbach, Nikolas Tezak, Robert Smith, Peter Karalekas, Anthony Polloreno, Sohaib Alam, Gavin Crooks, and Marcus Da Silva. Quantum kitchen sinks: An algorithm for machine learning on near-term quantum computers. Bulletin of the American Physical Society, 2019. 4 Discussion 13

    work page 2019

  13. [13]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Van- derplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine Learning in Python . Journal of Machine Learning Research, 12:2825–2830, 2011

    work page 2011

  14. [14]

    Random quantum circuits are approximate 2-designs

    Aram W Harrow and Richard A Low. Random quantum circuits are approximate 2-designs. Communications in Mathematical Physics , 291(1):257–302, 2009

    work page 2009

  15. [15]

    A variational eigenvalue solver on a photonic quantum processor

    Alberto Peruzzo, Jarrod McClean, Peter Shadbolt, Man-Hong Yung, Xiao-Qi Zhou, Peter J Love, Al´ an Aspuru-Guzik, and Jeremy L Obrien. A variational eigenvalue solver on a photonic quantum processor. Na- ture communications, 5:4213, 2014

    work page 2014

  16. [16]

    http://rodrigob.github.io/are we there yet/build/classification datasets results.html#4d4e495354

    work page

  17. [17]

    Lively quantum walks on cycles

    Przemys law Sadowski, Jaros law Adam Miszczak, and Mateusz Os- taszewski. Lively quantum walks on cycles. Journal of Physics A: Mathematical and Theoretical, 49(37):375302, 2016

    work page 2016

  18. [18]

    Various methods of optimiz- ing control pulses for quantum systems with decoherence

    Lukasz Pawela and Przemys law Sadowski. Various methods of optimiz- ing control pulses for quantum systems with decoherence. Quantum Information Processing, 15(5):1937–1953, 2016. 14 Machine Learning Kernel Method from a Quantum Generative Model 0.960 0.965 0.970 0.975 0.980 0.985 0.990 0.995 1.000 score 0 1 2 3 4 5 6 7 8count 0.960 0.965 0.970 0.975 0.98...

    work page 1937