pith. sign in

arxiv: 2405.08319 · v2 · submitted 2024-05-14 · 🪐 quant-ph

Measurement-based quantum machine learning

Luis Mantilla Calder\'on , Robert Raussendorf , Polina Feldmann , Dmytro Bondarenko This is my paper

Pith reviewed 2026-05-24 00:45 UTC · model grok-4.3

classification 🪐 quant-ph
keywords quantum machine learningmeasurement-based quantum computingquantum neural networkmultiple-triangle ansatzMBQC neuronsquantum kernelGKP qubitsuniversal ansatz
0
0 comments X

The pith

The multiple-triangle ansatz assembles a universal quantum neural network from MBQC neurons with bias engineering and tunable entanglement.

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

The paper introduces the multiple-triangle ansatz, or MuTA, to create quantum neural networks that operate natively in the measurement-based quantum computation model rather than the circuit model. These networks are built from basic MBQC neurons that support bias engineering, monotonic expressivity, tunable entanglement, and scalable training. Numerical tests demonstrate that MuTA learns a universal set of gates under noise, classifies quantum states, implements a quantum instrument, and classifies classical data via a tailored quantum kernel while respecting photonic GKP qubit constraints. This setup aims to bring quantum machine learning to devices that naturally support mid-circuit measurements and error correction.

Core claim

The central claim is that the multiple-triangle ansatz (MuTA) constitutes a universal quantum neural network assembled from MBQC neurons, featuring bias engineering, monotonic expressivity, tunable entanglement, and scalable training. Numerical demonstrations establish that MuTA can learn a universal set of gates in the presence of noise, serve as a quantum-state classifier and quantum instrument, classify classical data using a quantum kernel tailored to MuTA, and incorporate hardware constraints from photonic Gottesman-Kitaev-Preskill qubits.

What carries the argument

The multiple-triangle ansatz (MuTA), a universal quantum neural network assembled from MBQC neurons featuring bias engineering, monotonic expressivity, tunable entanglement, and scalable training.

If this is right

  • MuTA can learn a universal set of quantum gates even when noise is present.
  • MuTA functions as a quantum-state classifier and implements quantum instruments.
  • Classical data classification is possible using a quantum kernel designed specifically for MuTA.
  • Photonic GKP qubit hardware constraints can be directly incorporated into the network design.
  • Scalable training becomes feasible due to the monotonic expressivity property.

Where Pith is reading between the lines

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

  • MBQC's built-in mid-circuit measurements may allow hybrid classical-quantum training loops that reduce overall circuit depth compared with circuit-model approaches.
  • Tunable entanglement in MuTA could be used to study how entanglement structure affects generalization in quantum kernels.
  • The assembly from MBQC neurons suggests a path toward fault-tolerant quantum neural networks once MBQC error correction matures.
  • Monotonic expressivity may enable controlled increases in network capacity without encountering barren-plateau issues during training.

Load-bearing premise

The numerical demonstrations of learning under noise and with photonic GKP constraints will generalize beyond the specific simulations performed without post-hoc tuning that affects the reported performance.

What would settle it

An experiment or simulation in which MuTA fails to learn the target universal gate set at the reported noise levels or under the stated GKP constraints would falsify the claim of practical utility.

Figures

Figures reproduced from arXiv: 2405.08319 by Dmytro Bondarenko, Luis Mantilla Calder\'on, Polina Feldmann, Robert Raussendorf.

Figure 1
Figure 1. Figure 1: FIG. 1. Illustrations of the multiple-triangle ansatz (MuTA) architecture: (a) A MuTA for three qubits, showing input qubits [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. A MuTA layer implementing the same connectivity as [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Stability of MB-QML under noise affecting the train [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Learned surface to classify states of high ( [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Learning curve for learning a quantum instrument [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Classification of three different datasets using a SVM [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Embedding for kernel SVM classification of 2D [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. Learning curves for five runs of the greedy algo [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Learning curves for five runs of the DQN algo [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11. Effect of a depolarizing channel affecting the MuTA [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
read the original abstract

Quantum machine learning (QML) leverages quantum computing for classical inference, furnishes the processing of quantum data with machine-learning methods, and provides quantum algorithms adapted to noisy devices. Typically, QML proposals are framed in terms of the circuit model of quantum computation. The alternative measurement-based quantum computing (MBQC) paradigm can exhibit lower circuit depths, is naturally compatible with classical co-processing of mid-circuit measurements, and offers a promising avenue towards error correction. Despite significant progress on MBQC devices, QML in terms of MBQC has been hardly explored. We propose the multiple-triangle ansatz (MuTA), a universal quantum neural network assembled from MBQC neurons featuring bias engineering, monotonic expressivity, tunable entanglement, and scalable training. We numerically demonstrate that MuTA can learn a universal set of gates in the presence of noise, a quantum-state classifier, as well as a quantum instrument, and classify classical data using a quantum kernel tailored to MuTA. Finally, we incorporate hardware constraints imposed by photonic Gottesman-Kitaev-Preskill qubits. Our framework lays the foundation for versatile quantum neural networks native to MBQC, allowing to explore MBQC-specific algorithmic advantages and QML on MBQC devices.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes the multiple-triangle ansatz (MuTA), a universal quantum neural network assembled from MBQC neurons that incorporates bias engineering, monotonic expressivity, tunable entanglement, and scalable training. It numerically demonstrates MuTA's ability to learn a universal set of gates under noise, perform quantum-state classification, learn a quantum instrument, classify classical data via a MuTA-tailored quantum kernel, and incorporate photonic GKP qubit constraints.

Significance. If the numerical results prove robust, the work establishes the first systematic framework for measurement-based quantum machine learning. It opens exploration of MBQC-specific advantages such as reduced circuit depth, native classical co-processing of mid-circuit measurements, and compatibility with error correction, while providing a concrete ansatz for photonic hardware constraints.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (numerical demonstrations): the claims of success on gate learning, state classification, and instrument tasks under noise are stated without any quantitative metrics, baselines, error bars, or exclusion criteria. This absence prevents verification of the central performance claims and leaves the generalization assumption untested.
  2. [§4] §4 (GKP and noise simulations): the reported performance under specific noise models and GKP constraints lacks sensitivity analysis or ablation over modest changes in hyperparameters, noise strength, or circuit size. Without this, it is impossible to distinguish robust advantages from simulation-specific tuning.
minor comments (2)
  1. [MuTA definition] Clarify the precise definition of 'monotonic expressivity' and 'bias engineering' with explicit equations or pseudocode in the ansatz construction section.
  2. Add a table comparing MuTA resource requirements (qubits, measurements, classical post-processing) against standard circuit-model QNNs for the demonstrated tasks.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We address the two major comments below and commit to revisions that strengthen the quantitative presentation of our numerical results while preserving the manuscript's core contributions.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (numerical demonstrations): the claims of success on gate learning, state classification, and instrument tasks under noise are stated without any quantitative metrics, baselines, error bars, or exclusion criteria. This absence prevents verification of the central performance claims and leaves the generalization assumption untested.

    Authors: We agree that the absence of explicit quantitative metrics, statistical error bars, baselines, and exclusion criteria in the current text limits verifiability. In the revised manuscript we will add tables reporting average fidelities or accuracies with standard deviations over multiple independent training runs (e.g., 10–20 seeds), include simple baselines such as random guessing or non-entangling circuits, and state any data-exclusion rules. These additions will be placed in §4 and referenced from the abstract. revision: yes

  2. Referee: [§4] §4 (GKP and noise simulations): the reported performance under specific noise models and GKP constraints lacks sensitivity analysis or ablation over modest changes in hyperparameters, noise strength, or circuit size. Without this, it is impossible to distinguish robust advantages from simulation-specific tuning.

    Authors: The referee correctly identifies that the present simulations are performed at fixed hyperparameter and noise values. We will perform and report a limited sensitivity study in the revision: for the gate-learning and state-classification tasks we will vary depolarizing noise strength by ±20 % around the reported values and test one additional circuit depth; results will be summarized in new panels or a supplementary table. Full ablation across all hyperparameters remains computationally intensive, but the targeted checks will address the core concern of robustness versus tuning. revision: partial

Circularity Check

0 steps flagged

No significant circularity; MuTA introduced as independent construction with numerical demonstrations

full rationale

The paper proposes MuTA as a new ansatz assembled from MBQC neurons, claiming features such as bias engineering, monotonic expressivity, tunable entanglement, and scalable training without reducing these to prior fitted quantities or self-citations. Numerical demonstrations of gate learning under noise, state classification, instrument learning, and GKP constraints are presented as simulation results rather than predictions forced by construction from the same data. No load-bearing steps in the provided abstract or described claims equate outputs to inputs via self-definition, renaming, or imported uniqueness theorems. The derivation chain remains self-contained with independent architectural content.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The proposal rests on the established MBQC model and standard quantum mechanics; MuTA itself is the primary new construct introduced without external independent evidence beyond the reported simulations.

axioms (1)
  • domain assumption The measurement-based quantum computing paradigm provides a valid and universal model for quantum computation
    All constructions and demonstrations are framed inside MBQC.
invented entities (1)
  • MuTA (multiple-triangle ansatz) no independent evidence
    purpose: Universal quantum neural network native to MBQC with listed engineering features
    Newly proposed structure whose performance is shown only via the paper's own numerical tests.

pith-pipeline@v0.9.0 · 5747 in / 1315 out tokens · 46142 ms · 2026-05-24T00:45:19.463988+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

79 extracted references · 79 canonical work pages · 3 internal anchors

  1. [1]

    a non-entangling gate between wiresiandj∈Jif Ci,j is measured atα j ∈ {0, π},

    work page

  2. [2]

    Here, we call a gateGentangling for wiresiandjiff there exists somen-qubit state|ψ⟩=|ψ i⟩ ⊗ |ψj⟩ ⊗ |ψrest⟩ such thatG|ψ⟩reduced to the(i, j)-subspace is entangled

    an entangling gate between wiresiandj∈JifC i,j is measured atα j /∈ {0, π}. Here, we call a gateGentangling for wiresiandjiff there exists somen-qubit state|ψ⟩=|ψ i⟩ ⊗ |ψj⟩ ⊗ |ψrest⟩ such thatG|ψ⟩reduced to the(i, j)-subspace is entangled. Proof.In Appendix B, see Eq. (B2) and the subsequent paragraph, we translate a general MuTA layer into a 4 T1,2 (a) 0...

    work page

  3. [3]

    Adding disconnected wires

    work page

  4. [4]

    CanQuEST

    Adding admissible triangles to a layerlof widthn and inserting an additional layerl ′ of widthnbe- tween layerslandl+ 1. Layerl ′ can have any in- tralayer connectivity. It takes all outputs of layerl as inputs. The interlayer connectivity between lay- ersl ′ andl+ 1is inherited from the initial connec- tivity betweenlandl+ 1. Proof.For weak monotonicity,...

    work page 2000

  5. [5]

    Raussendorf and H

    R. Raussendorf and H. J. Briegel, Physical review letters 86, 5188 (2001)

    work page 2001

  6. [6]

    Raussendorf, D

    R. Raussendorf, D. E. Browne, and H. J. Briegel, Physical review A68, 022312 (2003)

    work page 2003

  7. [7]

    Raussendorf and T.-C

    R. Raussendorf and T.-C. Wei, Annual Review of Con- densed Matter Physics3, 239 (2012)

    work page 2012

  8. [8]

    Duan and R

    L.-M. Duan and R. Raussendorf, Phys. Rev. Lett.95, 080503 (2005)

    work page 2005

  9. [9]

    Chen, L.-C

    S. Chen, L.-C. Peng, Y.-P. Guo, X.-M. Gu, X. Ding, R.- Z. Liu, J.-Y. Zhao, X. You, J. Qin, Y.-F. Wang, Y.-M. He, J. J. Renema, Y.-H. Huo, H. Wang, C.-Y. Lu, and J.-W. Pan, Phys. Rev. Lett.132, 130603 (2024)

    work page 2024

  10. [10]

    Lib and Y

    O. Lib and Y. Bromberg, Nat. Phot.18, 1218 (2024)

    work page 2024

  11. [11]

    Yokoyama, R

    S. Yokoyama, R. Ukai, S. C. Armstrong, C. Sornphiphat- phong, T. Kaji, S. Suzuki, J.-i. Yoshikawa, H. Yonezawa, N. C. Menicucci, and A. Furusawa, Nat. Phot.7, 982 (2013)

    work page 2013

  12. [12]

    Mandel, M

    O. Mandel, M. Greiner, A. Widera, T. Rom, T. W. H¨ ansch, and I. Bloch, Nature425, 937 (2003)

    work page 2003

  13. [13]

    Broadbent and E

    A. Broadbent and E. Kashefi, Theoretical computer sci- ence410, 2489 (2009)

    work page 2009

  14. [14]

    Raussendorf, S

    R. Raussendorf, S. Bravyi, and J. Harrington, Phys. Rev. A71, 062313 (2005)

    work page 2005

  15. [15]

    Raussendorf, J

    R. Raussendorf, J. Harrington, and K. Goyal, Annals of Physics321, 2242 (2006)

    work page 2006

  16. [16]

    Rossi, L

    M. Rossi, L. Asproni, D. Caputo, S. Rossi, A. Cusinato, R. Marini, A. Agosti, and M. Magagnini, Using Shor’s algorithm on near term quantum computers: a reduced version (2021), arXiv:2112.12647 [quant-ph]

    work page arXiv 2021

  17. [17]

    D. Malz, G. Styliaris, Z.-Y. Wei, and J. I. Cirac, Physical Review Letters132, 10.1103/PhysRevLett.132.040404 (2024)

    work page doi:10.1103/physrevlett.132.040404 2024

  18. [18]

    Biamonte, P

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, Nature549, 195 (2017)

    work page 2017

  19. [19]

    Abbas, D

    A. Abbas, D. Sutter, C. Zoufal, A. Lucchi, A. Figalli, and S. Woerner, Nature Computational Science1, 403 (2021)

    work page 2021

  20. [20]

    I. Cong, S. Choi, and M. D. Lukin, Nature Physics15, 1273 (2019)

    work page 2019

  21. [21]

    K. Beer, D. Bondarenko, T. Farrelly, T. J. Osborne, R. Salzmann, D. Scheiermann, and R. Wolf, Nat Com- mun11, 808 (2020)

    work page 2020

  22. [22]

    Schuld and N

    M. Schuld and N. Killoran, Physical review letters122, 040504 (2019)

    work page 2019

  23. [23]

    Tilly, H

    J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth,et al., Physics Reports986, 1 (2022)

    work page 2022

  24. [24]

    J¨ ager and R

    J. J¨ ager and R. V. Krems, Nature Communications14, 576 (2023)

    work page 2023

  25. [25]

    Y. Liu, S. Arunachalam, and K. Temme, Nature Physics 17, 1013 (2021)

    work page 2021

  26. [26]

    R. R. Ferguson, L. Dellantonio, A. Al Balushi, K. Jansen, W. D¨ ur, and C. A. Muschik, Physical review letters126, 220501 (2021)

    work page 2021

  27. [27]

    Z. Qin, X. Li, Y. Zhou, S. Zhang, R. Li, C. Du, and Z. Xiao, New Journal of Physics26, 073040 (2024)

    work page 2024

  28. [28]

    Majumder, M

    A. Majumder, M. Krumm, T. Radkohl, L. J. Fiderer, H. P. Nautrup, S. Jerbi, and H. J. Briegel, Phys. Rev. A 110, 062616 (2024)

    work page 2024

  29. [29]

    D. E. Browne, E. Kashefi, M. Mhalla, and S. Perdrix, New Journal of Physics9, 250 (2007)

    work page 2007

  30. [30]

    Danos and E

    V. Danos and E. Kashefi, Physical Review A74, 052310 (2006)

    work page 2006

  31. [31]

    De Beaudrap, Physical Review A77, 022328 (2008)

    N. De Beaudrap, Physical Review A77, 022328 (2008)

    work page 2008

  32. [32]

    Mhalla and S

    M. Mhalla and S. Perdrix, inAutomata, Languages and Programming: 35th International Colloquium, ICALP 2008, Reykjavik, Iceland, July 7-11, 2008, Proceedings, Part I 35(Springer, 2008) pp. 857–868

    work page 2008

  33. [33]

    Goodfellow, Y

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

    work page 2016

  34. [34]

    Zhang, Z

    A. Zhang, Z. C. Lipton, M. Li, and A. J. Smola,Dive into Deep Learning(Cambridge University Press, 2023) https://D2L.ai

    work page 2023

  35. [35]

    Nair and G

    V. Nair and G. E. Hinton, inProceedings of the 27th international conference on machine learning (ICML-10) (2010) pp. 807–814

    work page 2010

  36. [36]

    A. J. da Silva, T. B. Ludermir, and W. R. de Oliveira, Neural Networks76, 55 (2016)

    work page 2016

  37. [37]

    Benedetti, E

    M. Benedetti, E. Lloyd, S. Sack, and M. Fiorentini, Quantum Science and Technology4, 043001 (2019)

    work page 2019

  38. [38]

    Danos, E

    V. Danos, E. Kashefi, and P. Panangaden, Physical Re- view A72, 064301 (2005)

    work page 2005

  39. [39]

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

    work page 2010

  40. [40]

    M. J. Bremner, C. M. Dawson, J. L. Dodd, A. Gilchrist, A. W. Harrow, D. Mortimer, M. A. Nielsen, and T. J. Osborne, Physical review letters89, 247902 (2002)

    work page 2002

  41. [41]

    Goyal, A

    K. Goyal, A. McCauley, and R. Raussendorf, Phys. Rev. A74, 032318 (2006)

    work page 2006

  42. [42]

    M. Hein, J. Eisert, and H. J. Briegel, Phys. Rev. A69, 062311 (2004)

    work page 2004

  43. [43]

    J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, and H. Neven, Nature communications9, 4812 (2018)

    work page 2018

  44. [44]

    Bondarenko and P

    D. Bondarenko and P. Feldmann, Physical review letters 124, 130502 (2020)

    work page 2020

  45. [45]

    Torrontegui and J

    E. Torrontegui and J. J. Garc´ ıa-Ripoll, Europhysics Let- ters125, 30004 (2019)

    work page 2019

  46. [46]

    D. P. Kingma and J. Ba, Adam: A method for stochastic optimization (2017), arXiv:1412.6980 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  47. [47]

    Poland, K

    K. Poland, K. Beer, and T. J. Osborne, arXiv preprint arXiv:2003.14103 10.48550/arXiv.2003.14103 (2020)

    work page doi:10.48550/arxiv.2003.14103 2003

  48. [48]

    Ragone, B

    M. Ragone, B. N. Bakalov, F. Sauvage, A. F. Kemper, C. Ortiz Marrero, M. Larocca, and M. Cerezo, Nature Communications15, 7172 (2024)

    work page 2024

  49. [49]

    Schatzki, M

    L. Schatzki, M. Larocca, Q. T. Nguyen, F. Sauvage, and M. Cerezo, Theoretical guarantees for permutation- equivariant quantum neural networks (2022), arXiv:2210.09974 [quant-ph]. 11

    work page arXiv 2022

  50. [50]

    Zhou and X

    T. Zhou and X. Chen, Physical Review E99, 052212 (2019)

    work page 2019

  51. [51]

    Garc´ ıa-Mart´ ın, M

    D. Garc´ ıa-Mart´ ın, M. Larocca, and M. Cerezo, Physical Review Research6, 013295 (2024)

    work page 2024

  52. [52]

    C. M. Bishop, Springer2, 1122 (2006)

    work page 2006

  53. [53]

    C. L. Degen, F. Reinhard, and P. Cappellaro, Reviews of modern physics89, 035002 (2017)

    work page 2017

  54. [54]

    Giovannetti, S

    V. Giovannetti, S. Lloyd, and L. Maccone, Nature pho- tonics5, 222 (2011)

    work page 2011

  55. [55]

    M. C. Caro, H.-Y. Huang, N. Ezzell, J. Gibbs, A. T. Sornborger, L. Cincio, P. J. Coles, and Z. Holmes, Nature Communications14, 3751 (2023)

    work page 2023

  56. [56]

    Li and S

    N. Li and S. Luo, Physical Review A88, 014301 (2013)

    work page 2013

  57. [57]

    Feldmann,Generalized Quantum Phase Transitions for Quantum-State Engineering in Spinor Bose-Einstein Condensates, Ph.D

    P. Feldmann,Generalized Quantum Phase Transitions for Quantum-State Engineering in Spinor Bose-Einstein Condensates, Ph.D. thesis, Gottfried Wilhelm Leibniz Universit¨ at Hannover (2021)

    work page 2021

  58. [58]

    T´ oth and D

    G. T´ oth and D. Petz, Physical Review A—Atomic, Molecular, and Optical Physics87, 032324 (2013)

    work page 2013

  59. [59]

    Gudder, Quanta12, 27 (2023)

    S. Gudder, Quanta12, 27 (2023)

    work page 2023

  60. [60]

    Havl´ ıˇ cek, A

    V. Havl´ ıˇ cek, A. D. C´ orcoles, K. Temme, A. W. Harrow, A. Kandala, J. M. Chow, and J. M. Gambetta, Nature 567, 209 (2019)

    work page 2019

  61. [61]

    Suzuki, H

    Y. Suzuki, H. Yano, Q. Gao, S. Uno, T. Tanaka, M. Akiyama, and N. Yamamoto, Quantum Machine In- telligence2, 1 (2020)

    work page 2020

  62. [62]

    Gentinetta, D

    G. Gentinetta, D. Sutter, C. Zoufal, B. Fuller, and S. Wo- erner, in2023 IEEE International Conference on Quan- tum Computing and Engineering (QCE), Vol. 1 (IEEE,

    work page

  63. [63]

    Gottesman, A

    D. Gottesman, A. Kitaev, and J. Preskill, Physical Re- view A64, 012310 (2001)

    work page 2001

  64. [64]

    Knill, R

    E. Knill, R. Laflamme, and G. J. Milburn, Nature409, 46 (2001)

    work page 2001

  65. [65]

    P. J. Shadbolt, M. R. Verde, A. Peruzzo, A. Politi, A. Laing, M. Lobino, J. C. Matthews, M. G. Thompson, and J. L. O’Brien, Nature Photonics6, 45 (2012)

    work page 2012

  66. [66]

    Fault-Tolerant Postselected Quantum Computation: Schemes

    E. Knill, Fault-tolerant postselected quantum computa- tion: Schemes (2004), arXiv:quant-ph/0402171 [quant- ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2004

  67. [67]

    Bravyi and J

    S. Bravyi and J. Haah, Physical Review A86, 052329 (2012)

    work page 2012

  68. [68]

    V. Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Ve- ness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski,et al., nature518, 529 (2015)

    work page 2015

  69. [69]

    Gottesman and I

    D. Gottesman and I. L. Chuang, Nature402, 390 (1999)

    work page 1999

  70. [70]

    Bondarenko, R

    D. Bondarenko, R. Salzmann, and V.-S. Schmiesing, arXiv preprint arXiv:2301.08167 (2023)

    work page arXiv 2023

  71. [71]

    W. F. Stinespring, Proceedings of the American Mathe- matical Society6, 211 (1955)

    work page 1955

  72. [72]

    Kidger and T

    P. Kidger and T. Lyons, Universal approximation with deep narrow networks, arXiv:1905.08539 (2020), arXiv:1905.08539 [cs.LG]

    work page arXiv 1905

  73. [73]

    Z. Lu, H. Pu, F. Wang, Z. Hu, and L. Wang, The expres- sive power of neural networks: A view from the width, arXiv:1709.02540 (2017), arXiv:1709.02540 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  74. [74]

    Perekrestenko, S

    D. Perekrestenko, S. M¨ uller, and H. B¨ olcskei, inInterna- tional Conference on Machine Learning(PMLR, 2020) pp. 7610–7619

    work page 2020

  75. [75]

    Glorot, A

    X. Glorot, A. Bordes, and Y. Bengio, inProceedings of the fourteenth international conference on artificial intel- ligence and statistics(JMLR Workshop and Conference Proceedings, 2011) pp. 315–323. 12 Appendix A: Monotonic expressivity of classical neurons When designing a model, it is very convenient if in- creasing its complexity leads to increased cap...

    work page 2011

  76. [76]

    We call such a path anf-path

    Repeatedly applying the flow functionfto a proper input qubititraces a path fromito a proper output qubit. We call such a path anf-path. Proof:ftraces a path becausef(j)is a neighbor ofj∀j; this path is finite becausef(j)> jand the number of qubits is finite; the final qubit of the path must be an output qubit because otherwise we could prolong the path b...

    work page

  77. [77]

    Proof: iff(i) =f(j)fori̸=jthenjis a neighbor off(i)such thatj > i, andiis a neighbor off(j) such thati > j, which is a contradiction

    Differentf-paths do not intersect. Proof: iff(i) =f(j)fori̸=jthenjis a neighbor off(i)such thatj > i, andiis a neighbor off(j) such thati > j, which is a contradiction

    work page

  78. [78]

    Qubits j∈I∩Odo not lie onf-paths

    Every qubitj /∈I∩Olies on anf-path. Qubits j∈I∩Odo not lie onf-paths. Proof: applyingfton− |I∩O|proper input qubits defines paths ending inn− |I∩O|different proper output qubits; ifj /∈I∩Odoes not lie on any of these paths, repeatedly applyingfmust trace a path 13 fromjto another proper output qubit; however, no other proper output qubits exist;j∈I∩Ocanno...

    work page

  79. [79]

    e., within a single f-path,Ghas no edges between non-consecutive qubits

    Allf-paths are induced paths; i. e., within a single f-path,Ghas no edges between non-consecutive qubits. Proof: letj < kbe non-consecutive qubits in anf- path; then there existsk ′ > jsuch thatk=f(k ′); ifjis a neighbor ofkthenj > k ′, which is a con- tradiction. Hence, any|G⟩consists ofn− |I∩O|interconnected 1-dimensional cluster states connected to|I∩O...

    work page