pith. sign in

arxiv: 2408.12739 · v2 · submitted 2024-08-22 · 🪐 quant-ph · cs.LG· stat.ML

Quantum Convolutional Neural Networks are Effectively Classically Simulable

Pablo Bermejo , Paolo Braccia , Manuel S. Rudolph , Zo\"e Holmes , Lukasz Cincio , M. Cerezo This is my paper

Pith reviewed 2026-05-23 22:02 UTC · model grok-4.3

classification 🪐 quant-ph cs.LGstat.ML
keywords quantum convolutional neural networksclassical simulabilityPauli shadowsquantum machine learninglow-bodyness observablesphases of matter classificationvariational quantum algorithms
0
0 comments X

The pith

Randomly initialized QCNNs only access low-bodyness measurements and can be classically simulated on standard benchmarks.

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

The paper shows that randomly initialized QCNNs are confined to processing information from low-bodyness measurements on input states. Standard benchmark datasets for tasks like phase classification are exactly those solvable from the same low-bodyness information. A classical algorithm using Pauli shadows therefore reproduces the QCNN output efficiently. This simulation is demonstrated on systems up to 1024 qubits. The result implies that reported QCNN performance may reflect the choice of simple, simulable problems rather than any intrinsic quantum capability.

Core claim

When randomly initialized, QCNNs can only operate on the information encoded in low-bodyness measurements of their input states. They are commonly benchmarked on locally-easy datasets whose states are precisely classifiable by the information encoded in these low-bodyness observables subspace. The QCNN's action on this subspace can be efficiently classically simulated by a classical algorithm equipped with Pauli shadows on the dataset, with explicit demonstrations on up to 1024 qubits for phases of matter classification.

What carries the argument

The low-bodyness measurement subspace, which both restricts randomly initialized QCNN operation and permits efficient Pauli-shadow classical simulation of its outputs.

If this is right

  • QCNN performance on common phases-of-matter tasks can be replicated by a classical shadow algorithm up to at least 1024 qubits.
  • Heuristic success of QCNNs on existing benchmarks follows from the datasets lying inside the low-bodyness subspace.
  • The same shadow-based argument extends to other variational quantum architectures that share the low-bodyness restriction.
  • Non-trivial datasets containing information outside the low-bodyness subspace are required to test whether QCNNs can exceed classical simulation.

Where Pith is reading between the lines

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

  • Many current QML benchmark tasks may be inadequate for revealing quantum advantage because they are solvable by low-body classical methods.
  • New datasets built around high-body or long-range correlations would serve as a direct test of whether the low-bodyness limitation can be overcome.
  • If other variational models also begin in a low-bodyness regime, similar classical shadow simulations could be applied to them without further modification.

Load-bearing premise

Randomly initialized QCNNs stay restricted to the low-bodyness measurement subspace and the benchmark datasets are exactly those classifiable from information inside that same subspace.

What would settle it

An experiment in which a randomly initialized QCNN achieves high accuracy on a phases-of-matter dataset that requires observables of body greater than two, while the corresponding Pauli-shadow classical simulator fails to match that accuracy.

Figures

Figures reproduced from arXiv: 2408.12739 by Lukasz Cincio, Manuel S. Rudolph, M. Cerezo, Pablo Bermejo, Paolo Braccia, Zo\"e Holmes.

Figure 1
Figure 1. Figure 1: Schematic representation of our main results. We conceptualize the success of QCNN as a consequence of two facts: (1) When randomly initialized, they operate on a polynomially-sized subspace of low-bodyness observables, (2) They are benchmarked on locally-easy datasets that are clas￾sifiable via the information encoded in low-bodyness measure￾ments. The combination of these two facts allows us to show that… view at source ↗
Figure 2
Figure 2. Figure 2: QCNN architecture. (a) A QCNN is composed of alternating convolutional and pooling layers. In the con￾volutional layers, information is usually being processed by parametrized quantum gates. In the pooling layers, the dimen￾sion of the QCNN feature space is reduced by tracing out or measuring qubits. By design, QCNNs have a depth that only scales logarithmically with the number of qubits n, and the measure… view at source ↗
Figure 3
Figure 3. Figure 3: Bond-Alternating XXX model. a) Phase diagram for the Hamiltonian in Eq. (6). b) Test classification accuracy for the simulated QCNN acting only on the low-bodyness op￾erator subspace. We show the accuracy as a function of the number of training points and Pauli classical shadows on each state of the dataset. information from the operator space is needed for the clas￾sification, we remark that only the firs… view at source ↗
Figure 4
Figure 4. Figure 4: Haldane Chain. a) Phase diagram for the Hamilto￾nian in Eq. (7). b) Test classification accuracy for the simulated QCNN acting only on the low-bodyness operator subspace. We show the accuracy as a function of the number of training points and Pauli classical shadows on each state of the dataset. 2. Haldane Chain Next, we will consider the one-dimensional Haldane chain model [108] whose Hamiltonian reads H … view at source ↗
Figure 5
Figure 5. Figure 5: ANNNI model. a) Phase diagram for the Hamiltonian in Eq. (8). The phases are: (I) ferromagnetic, (II) paramagnetic, (III) floating, (IV) antiphase. b) Predicted phase diagram when training the simulated QCNN acting only on the low-bodyness operator subspace. The model is trained on with 200 and 20 states. The crosses mark the training points, while the circle the test ones. Blue circles means correct phase… view at source ↗
Figure 6
Figure 6. Figure 6: Cluster model. a) Phase diagram for the Hamiltonian in Eq. (9). The phases are: (I) Haldane, (II) ferromagnetic, (III) anti-ferromagnetic phase, and (IV) trivial. b) Predicted phase diagram when training the simulated QCNN acting only on the low-bodyness operator subspace. The model is trained on with 200 and 20 states. The crosses mark the training points, while the circle the test ones. Blue circles mean… view at source ↗
Figure 7
Figure 7. Figure 7: Schematic representation of the proof technique. In this figure we show a QCNN architecture (left) and the spread of the vectorized measurement operator as it propagates through the P gates of the first 2 layers of the QCNN. The initial state is given by |s⟩ on the first qubit and |i⟩ on the remaining qubits. Here, we also include the coefficient a defining the contribution of each operator to the k-puriti… view at source ↗
Figure 8
Figure 8. Figure 8: Average contribution per bodyness for different problem sizes. In this figure we plot 1 3k( n k) Eθ  p (k) Φ † θ (O)  for a system of n = 128 qubits. This quantity determines the average contribution of a given Pauli with bodyness k on the Heisenberg evolved measurement operator. Appendix B: Classical simulability of QCNNs Having realized that the success of QCNNs arises from the fact that they are initi… view at source ↗
Figure 9
Figure 9. Figure 9: Schematic comparison between Pauli Propagation (PP) and Pauli Propagation Surrogate (PPS) methods. Both methods are based on propagating Pauli operators through a quantum circuit which will in general create new Pauli operators via splitting of paths. If we employ a breadth-first approach to the propagation, we can identify identical Pauli operators and merge them. The splitting and merging pattern depicte… view at source ↗
read the original abstract

Quantum Convolutional Neural Networks (QCNNs) are widely regarded as a promising model for Quantum Machine Learning (QML). In this work we tie their heuristic success to two facts. First, that when randomly initialized, they can only operate on the information encoded in low-bodyness measurements of their input states. And second, that they are commonly benchmarked on "locally-easy'' datasets whose states are precisely classifiable by the information encoded in these low-bodyness observables subspace. We further show that the QCNN's action on this subspace can be efficiently classically simulated by a classical algorithm equipped with Pauli shadows on the dataset. Indeed, we present a shadow-based simulation of QCNNs on up-to $1024$ qubits for phases of matter classification. Our results can then be understood as highlighting a deeper symptom of QML: Models could only be showing heuristic success because they are benchmarked on simple problems, for which their action can be classically simulated. This insight points to the fact that non-trivial datasets are a truly necessary ingredient for moving forward with QML. To finish, we discuss how our results can be extrapolated to classically simulate other architectures.

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 claims that QCNNs succeed heuristically because random initialization restricts them to low-bodyness measurements and because common benchmark datasets (e.g., phases of matter) are classifiable from that subspace alone. It further asserts that QCNN action on this subspace is classically simulable via Pauli shadow tomography and demonstrates such a simulation on up to 1024 qubits. The work interprets this as evidence that QML heuristic success often reflects simple, classically simulable problems and calls for non-trivial datasets; it also discusses extrapolation to other architectures.

Significance. If the central claims hold, the result would be significant for QML evaluation practices by showing that certain architectures remain classically simulable on standard benchmarks even at large scale. The explicit shadow-based simulation reaching 1024 qubits constitutes a concrete, reproducible strength that quantifies the simulability claim for the phases-of-matter task.

major comments (2)
  1. [Abstract, paragraph 2] Abstract, paragraph 2: The restriction to the low-bodyness subspace is derived only for randomly initialized QCNNs. No argument or numerical check is supplied showing that gradient updates preserve membership in this subspace or that the trained circuit remains equivalent to a low-body observable; this gap is load-bearing for the claim that trained QCNNs are classically simulable.
  2. [Abstract, paragraph 2] Abstract, paragraph 2 and § on datasets: The statement that benchmark datasets are 'precisely classifiable by the information encoded in these low-bodyness observables' is asserted without a control experiment demonstrating that higher-weight observables cannot achieve higher accuracy on the same phases-of-matter data; this assumption directly supports the conclusion that the problems are classically simulable.
minor comments (2)
  1. [Introduction] Notation for 'low-bodyness' and 'low-body subspace' is used interchangeably without an explicit definition or reference to the precise operator support size; a short definition in the introduction would improve clarity.
  2. [Shadow simulation section] The shadow-tomography simulation section would benefit from an explicit statement of the number of shadows required and the error bounds used for the 1024-qubit experiments.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the scope of our claims regarding random initialization versus trained models and the sufficiency of low-bodyness for the benchmark datasets. We provide point-by-point responses below and indicate where revisions will be made to address the concerns.

read point-by-point responses

  1. Referee: The restriction to the low-bodyness subspace is derived only for randomly initialized QCNNs. No argument or numerical check is supplied showing that gradient updates preserve membership in this subspace or that the trained circuit remains equivalent to a low-body observable; this gap is load-bearing for the claim that trained QCNNs are classically simulable.

    Authors: The derivation is indeed presented for the random initialization case, as the low-bodyness follows directly from the initial circuit structure before any training. For the trained case, the claim of classical simulability assumes that the optimization does not take the model outside this subspace. While this is plausible given the local nature of QCNNs, we acknowledge the absence of an explicit argument or check in the current manuscript. In the revision, we will add a section providing a proof that the bodyness is preserved under gradient updates due to the convolutional architecture, along with numerical verification on small instances. This addresses the load-bearing aspect of the claim. revision: yes

  2. Referee: The statement that benchmark datasets are 'precisely classifiable by the information encoded in these low-bodyness observables' is asserted without a control experiment demonstrating that higher-weight observables cannot achieve higher accuracy on the same phases-of-matter data; this assumption directly supports the conclusion that the problems are classically simulable.

    Authors: We recognize that the assertion would be strengthened by an explicit control. The phases-of-matter datasets are chosen because they are known to be distinguishable by local properties, but we did not include a direct comparison with higher-body observables. We will incorporate a control experiment in the revised manuscript, training or evaluating classifiers using higher-weight Pauli observables on the same data to show that they do not yield improved accuracy, thereby confirming the sufficiency of the low-bodyness subspace. revision: yes

Circularity Check

0 steps flagged

No significant circularity; simulation applies standard shadow tomography to low-body subspace identified by QCNN structure.

full rationale

The paper's derivation identifies that randomly initialized QCNNs are restricted to low-bodyness observables and that common benchmark datasets are classifiable from that subspace, then applies Pauli shadow tomography (a standard external technique) to simulate the action on that subspace. No step reduces by construction to a fitted parameter, self-definition, or load-bearing self-citation chain; the central simulability result follows from the QCNN architecture plus known shadow methods without renaming or smuggling ansatzes. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The argument relies on standard quantum information assumptions (Pauli measurements, shadow tomography) without introducing new free parameters or postulated entities.

axioms (1)
  • standard math Standard properties of Pauli shadows and low-body observables in quantum information
    Invoked to justify the classical simulation procedure

pith-pipeline@v0.9.0 · 5755 in / 1246 out tokens · 77202 ms · 2026-05-23T22:02:14.201151+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 6 Pith papers

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

  1. Accelerating Inference for Multilayer Neural Networks with Quantum Computers

    quant-ph 2025-10 unverdicted novelty 7.0

    Quantum circuits for coherent multilayer neural network inference achieve quadratic to polylogarithmic speedups over classical methods depending on quantum data access models for inputs and weights.

  2. Tensor network surrogate models for variational quantum computation

    quant-ph 2026-04 unverdicted novelty 6.0

    Tensor network simulations act as effective surrogate models for training QAOA on large 2D lattices, overcoming limits of parameter transfer from small instances and remaining classically feasible with moderate bond d...

  3. Variational decision diagrams for quantum-inspired machine learning applications

    quant-ph 2025-02 unverdicted novelty 6.0

    The paper proposes variational decision diagrams (VDDs) for quantum state representation in QML and reports successful training without barren plateaus on transverse-field Ising and Heisenberg Hamiltonians.

  4. Benchmarking quantum devices beyond classical capabilities

    quant-ph 2025-02 unverdicted novelty 6.0

    Modified Quantum Volume test uses restricted universal circuits to directly determine heavy outputs without exponential classical simulation cost.

  5. Optimal quantum reservoir learning in proximity to universality

    quant-ph 2025-10 unverdicted novelty 5.0

    A tunable mixing parameter p in random quantum circuits controls the transition from classically simulable to expressive quantum reservoir dynamics via entanglement and nonstabilizer content.

  6. Iterative Quantum Feature Maps

    quant-ph 2025-06 unverdicted novelty 5.0

    IQFMs iteratively constructs deep quantum feature maps from shallow circuits via classical augmentation weights and contrastive layer-wise training, outperforming QCNNs on noisy quantum data and matching classical neu...

Reference graph

Works this paper leans on

138 extracted references · 138 canonical work pages · cited by 6 Pith papers · 2 internal anchors

  1. [1]

    In particular, we will see that simple Pauli clas- sical shadows are sufficient to extract the relevant features of the data

    Proving that datasets commonly used used in the lit- erature as heuristic evidence for QCNNs are locally- easy. In particular, we will see that simple Pauli clas- sical shadows are sufficient to extract the relevant features of the data

    work page

  2. [2]

    Illustrating the classical simulation of QCNNs, as well as the scalability of the proposed techniques (we simulate and train QCNNs with upto 1024 qubits)

    work page

  3. [3]

    full QCNNs

    Studying how the performance of the simulated QCNNs depends on the amount of measurements used during the shadow tomographic procedure. The simulations presented in this section are based on two techniques which are constructed to only process the information encoded in low-bodyness measurements of the input states. In particular, they accomplish this goa...

    work page

  4. [4]

    Heisenberg bond-alternating XXX model The Hamiltonian of the Heisenberg Bond-Alternating XXX model reads H = n−1X i=1 Ji (XiXi+1 + YiYi+1 + ZiZi+1) , (6) where Xi, Yi and Zi denotethePaulioperatoractingonthe i-thqubit, andwhere Ji = J1(J2) ⩾ 0for i = even(odd). In the thermodynamic limitn → ∞, the ground state of this model presents a phase transition con...

    work page

  5. [5]

    (7) Here, we take J > 0

    Haldane Chain Next, we will consider the one-dimensional Haldane chain model [108] whose Hamiltonian reads H = −J n−2X i=1 ZiXi+1Zi+2 −h1 nX i=1 Xi −h2 n−1X i=1 XiXi+1 . (7) Here, we take J > 0. As shown in Fig. 4 the model’s phase diagram is characterized by the two ratiosh1/J and h2/J. Inspecting the model in the thermodynamic limit, one can find out th...

    work page

  6. [6]

    (8) The phases of this model are classified in terms of the di- mensionless ratios κ = −J2/J1 and h = B/J1

    ANNNI model In this section, we consider the ANNNI model [109], which is described by the following Hamiltonian H = −J1 n−1X i=1 XiXi+1 − J2 n−2X i=1 XiXi+2 − B nX i=1 Zi . (8) The phases of this model are classified in terms of the di- mensionless ratios κ = −J2/J1 and h = B/J1. The phase diagram for this model is presented in Fig. 5. In particular, 9 we...

    work page

  7. [7]

    classical training and quantum deployment

    Cluster Hamiltonian To finish, we consider the cluster model defined by the following Hamiltonian H = nX i=1 (Zi − J1XiXi+1 − J2Xi−1ZiXi+1) . (9) Note that here we employ closed boundary conditions so that Xn+1 ≡ X1 and X−1 ≡ Xn. As shown in Fig. 6, the phase transitions are delimited by the ratios betweenJ1 and J2. Similarly to the previous ANNNI model, ...

    work page

  8. [8]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, and A. Courville,Deep Learning (MIT Press, 2016)

    work page 2016

  9. [9]

    LeCun, Y

    Y. LeCun, Y. Bengio, and G. Hinton, Deep learning, na- ture 521, 436 (2015)

    work page 2015

  10. [10]

    Schmidhuber, Deep learning in neural networks: An overview, Neural networks61, 85 (2015)

    J. Schmidhuber, Deep learning in neural networks: An overview, Neural networks61, 85 (2015)

    work page 2015

  11. [11]

    M. M. Bronstein, J. Bruna, T. Cohen, and P. Veličković, Geometric deep learning: Grids, groups, graphs, geodesics, and gauges, arXiv preprint arXiv:2104.13478 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  12. [12]

    Biamonte, P

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, Quantum machine learning, Na- ture 549, 195 (2017)

    work page 2017

  13. [13]

    Schuld and F

    M. Schuld and F. Petruccione,Supervised learning with quantum computers, Vol. 17 (Springer, 2018)

    work page 2018

  14. [14]

    Cerezo, A

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, and P. J. Coles, Variational quantum algo- rithms, Nature Reviews Physics3, 625–644 (2021)

    work page 2021

  15. [15]

    Bharti, A

    K. Bharti, A. Cervera-Lierta, T. H. Kyaw, T. Haug, S. Alperin-Lea, A. Anand, M. Degroote, H. Heimonen, J. S. Kottmann, T. Menke, et al., Noisy intermediate- scale quantum algorithms, Reviews of Modern Physics94, 015004 (2022)

    work page 2022

  16. [16]

    Cerezo, G

    M. Cerezo, G. Verdon, H.-Y. Huang, L. Cincio, and P. J. Coles, Challenges and opportunities in quantum machine learning, Nature Computational Science 10.1038/s43588- 022-00311-3 (2022)

    work page doi:10.1038/s43588- 2022

  17. [17]

    S. Endo, Z. Cai, S. C. Benjamin, and X. Yuan, Hybrid quantum-classical algorithms and quantum error mitiga- tion, Journal of the Physical Society of Japan90, 032001 (2021)

    work page 2021

  18. [18]

    Larocca, S

    M. Larocca, S. Thanasilp, S. Wang, K. Sharma, J. Bia- monte, P. J. Coles, L. Cincio, J. R. McClean, Z. Holmes, and M. Cerezo, A review of barren plateaus in variational quantum computing, arXiv preprint arXiv:2405.00781 (2024)

    work page arXiv 2024

  19. [19]

    J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, andH.Neven,Barrenplateausinquantumneuralnetwork training landscapes, Nature Communications9, 1 (2018)

    work page 2018

  20. [20]

    Cerezo, A

    M. Cerezo, A. Sone, T. Volkoff, L. Cincio, and P. J. Coles, Cost function dependent barren plateaus in shallow parametrized quantum circuits, Nature Communications 12, 1 (2021)

    work page 2021

  21. [21]

    Cerezo and P

    M. Cerezo and P. J. Coles, Higher order derivatives of quantum neural networks with barren plateaus, Quantum Science and Technology6, 035006 (2021)

    work page 2021

  22. [22]

    Arrasmith, Z

    A. Arrasmith, Z. Holmes, M. Cerezo, and P. J. Coles, Equivalence of quantum barren plateaus to cost concen- tration and narrow gorges, Quantum Science and Tech- nology 7, 045015 (2022)

    work page 2022

  23. [23]

    Ragone, B

    M. Ragone, B. N. Bakalov, F. Sauvage, A. F. Kemper, C. O. Marrero, M. Larocca, and M. Cerezo, A unified theory of barren plateaus for deep parametrized quantum circuits, arXiv preprint arXiv:2309.09342 (2023)

    work page arXiv 2023

  24. [24]

    Holmes, K

    Z. Holmes, K. Sharma, M. Cerezo, and P. J. Coles, Con- necting ansatz expressibility to gradient magnitudes and barren plateaus, PRX Quantum3, 010313 (2022)

    work page 2022

  25. [25]

    Bittel and M

    L. Bittel and M. Kliesch, Training variational quantum al- gorithmsisNP-hard,Phys.Rev.Lett. 127,120502(2021)

    work page 2021

  26. [26]

    Fontana, M

    E. Fontana, M. Cerezo, A. Arrasmith, I. Rungger, and P.J.Coles,Non-trivialsymmetriesinquantumlandscapes and their resilience to quantum noise, Quantum6, 804 (2022)

    work page 2022

  27. [27]

    E. R. Anschuetz and B. T. Kiani, Beyond barren plateaus: Quantum variational algorithms are swamped with traps, Nature Communications13, 7760 (2022). 12

    work page 2022

  28. [28]

    E. R. Anschuetz, Critical points in quantum generative models, International Conference on Learning Represen- tations (2022)

    work page 2022

  29. [29]

    Larocca, P

    M. Larocca, P. Czarnik, K. Sharma, G. Muraleedharan, P. J. Coles, and M. Cerezo, Diagnosing Barren Plateaus with Tools from Quantum Optimal Control, Quantum6, 824 (2022)

    work page 2022

  30. [30]

    P.Bermejo, B.Aizpurua,andR.Orús,Improvinggradient methods via coordinate transformations: Applications to quantum machine learning, Physical Review Research6, 023069 (2024)

    work page 2024

  31. [31]

    Gil-Fuster, C

    E. Gil-Fuster, C. Gyurik, A. Pérez-Salinas, and V. Dun- jko, On the relation between trainability and dequan- tization of variational quantum learning models, arXiv preprint arXiv:2406.07072 (2024)

    work page arXiv 2024

  32. [32]

    Pesah, M

    A. Pesah, M. Cerezo, S. Wang, T. Volkoff, A. T. Sorn- borger, and P. J. Coles, Absence of barren plateaus in quantum convolutional neural networks, Physical Review X 11, 041011 (2021)

    work page 2021

  33. [33]

    Khatri, R

    S. Khatri, R. LaRose, A. Poremba, L. Cincio, A. T. Sorn- borger, and P. J. Coles, Quantum-assisted quantum com- piling, Quantum3, 140 (2019)

    work page 2019

  34. [34]

    Zhao and X.-S

    C. Zhao and X.-S. Gao, Analyzing the barren plateau phe- nomenon in training quantum neural networks with the ZX-calculus, Quantum5, 466 (2021)

    work page 2021

  35. [35]

    Liu, L.-W

    Z. Liu, L.-W. Yu, L.-M. Duan, and D.-L. Deng, The pres- ence and absence of barren plateaus in tensor-network based machine learning, Physical Review Letters 129, 270501 (2022)

    work page 2022

  36. [36]

    Miao and T

    Q. Miao and T. Barthel, Isometric tensor network op- timization for extensive hamiltonians is free of barren plateaus, Physical Review A109, L050402 (2024)

    work page 2024

  37. [37]

    Monbroussou, J

    L. Monbroussou, J. Landman, A. B. Grilo, R. Kukla, and E. Kashefi, Trainability and expressivity of hamming- weight preserving quantum circuits for machine learning, arXiv preprint arXiv:2309.15547 (2023)

    work page arXiv 2023

  38. [38]

    S. Raj, I. Kerenidis, A. Shekhar, B. Wood, J. Dee, S. Chakrabarti, R. Chen, D. Herman, S. Hu, P. Minssen, et al., Quantum deep hedging, Quantum7, 1191 (2023)

    work page 2023

  39. [39]

    Fontana, D

    E. Fontana, D. Herman, S. Chakrabarti, N. Kumar, R. Yalovetzky, J. Heredge, S. Hari Sureshbabu, and M. Pistoia, The adjoint is all you need: Characteriz- ing barren plateaus in quantum ansätze, arXiv preprint arXiv:2309.07902 (2023)

    work page arXiv 2023

  40. [40]

    N. L. Diaz, D. García-Martín, S. Kazi, M. Larocca, and M. Cerezo, Showcasing a barren plateau the- ory beyond the dynamical lie algebra, arXiv preprint arXiv:2310.11505 (2023)

    work page arXiv 2023

  41. [41]

    M. T. West, J. Heredge, M. Sevior, and M. Usman, Prov- ably trainable rotationally equivariant quantum machine learning, PRX Quantum5, 030320 (2024)

    work page 2024

  42. [42]

    Zhang, L

    K. Zhang, L. Liu, M.-H. Hsieh, and D. Tao, Escaping from thebarrenplateauviagaussianinitializationsindeepvari- ational quantum circuits, inAdvances in Neural Informa- tion Processing Systems(2022)

    work page 2022

  43. [43]

    Park and N

    C.-Y. Park and N. Killoran, Hamiltonian variational ansatz without barren plateaus, Quantum8, 1239 (2024)

    work page 2024

  44. [44]

    Y. Wang, B. Qi, C. Ferrie, and D. Dong, Trainabil- ity enhancement of parameterized quantum circuits via reduced-domain parameter initialization, arXiv preprint arXiv:2302.06858 (2023)

    work page arXiv 2023

  45. [45]

    Larocca, F

    M. Larocca, F. Sauvage, F. M. Sbahi, G. Verdon, P. J. Coles, and M. Cerezo, Group-invariant quantum machine learning, PRX Quantum3, 030341 (2022)

    work page 2022

  46. [46]

    J. J. Meyer, M. Mularski, E. Gil-Fuster, A. A. Mele, F. Arzani, A. Wilms, and J. Eisert, Exploiting symmetry in variational quantum machine learning, PRX Quantum 4, 010328 (2023)

    work page 2023

  47. [47]

    Skolik, M

    A. Skolik, M. Cattelan, S. Yarkoni, T. Bäck, and V. Dun- jko,Equivariantquantumcircuitsforlearningonweighted graphs, npj Quantum Information9, 47 (2023)

    work page 2023

  48. [48]

    Ragone, Q

    M. Ragone, Q. T. Nguyen, L. Schatzki, P. Braccia, M. Larocca, F. Sauvage, P. J. Coles, and M. Cerezo, Rep- resentation theory for geometric quantum machine learn- ing, arXiv preprint arXiv:2210.07980 (2022)

    work page arXiv 2022

  49. [49]

    Q. T. Nguyen, L. Schatzki, P. Braccia, M. Ragone, P. J. Coles, F. Sauvage, M. Larocca, and M. Cerezo, Theory for equivariant quantum neural networks, PRX Quantum 5, 020328 (2024)

    work page 2024

  50. [50]

    Schatzki, M

    L. Schatzki, M. Larocca, Q. T. Nguyen, F. Sauvage, and M. Cerezo, Theoretical guarantees for permutation- equivariant quantum neural networks, npj Quantum In- formation 10, 12 (2024)

    work page 2024

  51. [51]

    Kieferova, O

    M. Kieferova, O. M. Carlos, and N. Wiebe, Quantum gen- erative training using rényi divergences, arXiv preprint arXiv:2106.09567 (2021)

    work page arXiv 2021

  52. [52]

    M. S. Rudolph, S. Lerch, S. Thanasilp, O. Kiss, S. Val- lecorsa, M. Grossi, and Z. Holmes, Trainability barriers and opportunities in quantum generative modeling, arXiv preprint arXiv:2305.02881 (2023)

    work page arXiv 2023

  53. [53]

    Letcher, S

    A. Letcher, S. Woerner, and C. Zoufal, Tight and effi- cient gradient bounds for parameterized quantum circuits, arXiv preprint arXiv:2309.12681 (2023)

    work page arXiv 2023

  54. [54]

    M. S. Rudolph, J. Miller, D. Motlagh, J. Chen, A. Acharya, and A. Perdomo-Ortiz, Synergistic pretrain- ing of parametrized quantum circuits via tensor networks, Nature Communications14, 8367 (2023)

    work page 2023

  55. [55]

    Cerezo, M

    M. Cerezo, M. Larocca, D. García-Martín, N. L. Diaz, P. Braccia, E. Fontana, M. S. Rudolph, P. Bermejo, A. Ijaz, S. Thanasilp,et al., Does provable absence of bar- ren plateaus imply classical simulability? or, why we need to rethink variational quantum computing, arXiv preprint arXiv:2312.09121 (2023)

    work page arXiv 2023

  56. [56]

    I. Cong, S. Choi, and M. D. Lukin, Quantum convolu- tional neural networks, Nature Physics15, 1273 (2019)

    work page 2019

  57. [57]

    T. Hur, L. Kim, and D. K. Park, Quantum convolutional neural network for classical data classification, Quantum Machine Intelligence4, 3 (2022)

    work page 2022

  58. [58]

    S. Oh, J. Choi, and J. Kim, A tutorial on quantum con- volutional neural networks (qcnn), in2020 International Conference on Information and Communication Technol- ogy Convergence (ICTC)(IEEE, 2020) pp. 236–239

    work page 2020

  59. [59]

    H. Baek, W. J. Yun, and J. Kim, Scalable quantum convo- lutional neural networks, arXiv preprint arXiv:2209.12372 13 (2022)

    work page arXiv 2022

  60. [60]

    L.-H.Gong, J.-J.Pei, T.-F.Zhang,andN.-R.Zhou,Quan- tum convolutional neural network based on variational quantum circuits, Optics Communications 550, 129993 (2024)

    work page 2024

  61. [61]

    J. Kim, J. Huh, and D. K. Park, Classical-to-quantum convolutional neural network transfer learning, Neuro- computing 555, 126643 (2023)

    work page 2023

  62. [62]

    Li, P.-C

    W. Li, P.-C. Chu, G.-Z. Liu, Y.-B. Tian, T.-H. Qiu, and S.-M. Wang, An image classification algorithm based on hybrid quantum classical convolutional neural network, Quantum Engineering2022, 5701479 (2022)

    work page 2022

  63. [63]

    Bokhan, A

    D. Bokhan, A. S. Mastiukova, A. S. Boev, D. N. Trub- nikov, and A. K. Fedorov, Multiclass classification us- ing quantum convolutional neural networks with hy- brid quantum-classical learning, Frontiers in Physics10, 1069985 (2022)

    work page 2022

  64. [64]

    Ovalle-Magallanes, D

    E. Ovalle-Magallanes, D. E. Alvarado-Carrillo, J. G. Avina-Cervantes, I. Cruz-Aceves, and J. Ruiz-Pinales, Quantum angle encoding with learnable rotation applied to quantum–classical convolutional neural networks, Ap- plied Soft Computing141, 110307 (2023)

    work page 2023

  65. [65]

    S. Y. Chang, M. Grossi, B. Le Saux, and S. Vallecorsa, Approximately equivariant quantum neural network for p4m group symmetries in images, in2023 IEEE Interna- tional Conference on Quantum Computing and Engineer- ing (QCE), Vol. 01 (2023) pp. 229–235

    work page 2023

  66. [66]

    F. Fan, Y. Shi, T. Guggemos, and X. X. Zhu, Hybrid quantum-classical convolutional neural net- work model for image classification, IEEE trans- actions on neural networks and learning systems 10.1109/TNNLS.2023.3312170 (2023)

    work page doi:10.1109/tnnls.2023.3312170 2023

  67. [67]

    Li, R.-G

    Y. Li, R.-G. Zhou, R. Xu, J. Luo, and W. Hu, A quantum deep convolutional neural network for image recognition, Quantum Science and Technology5, 044003 (2020)

    work page 2020

  68. [68]

    Y. Zeng, H. Wang, J. He, Q. Huang, and S. Chang, A multi-classification hybrid quantum neural network using an all-qubit multi-observable measurement strategy, En- tropy 24, 394 (2022)

    work page 2022

  69. [69]

    Sebastianelli, D

    A. Sebastianelli, D. A. Zaidenberg, D. Spiller, B. Le Saux, and S. L. Ullo, On circuit-based hybrid quantum neural networks for remote sensing imagery classification, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing15, 565 (2021)

    work page 2021

  70. [70]

    Chalumuri, R

    A. Chalumuri, R. Kune, S. Kannan, and B. Manoj, Quantum–classical image processing for scene classifica- tion, IEEE Sensors Letters6, 1 (2022)

    work page 2022

  71. [71]

    Zhang, J

    X. Zhang, J. Xia, X. Tan, X. Zhou, and T. Wang, Polsar image classification via learned superpixels and qcnn in- tegrating color features, Remote Sensing11, 1831 (2019)

    work page 2019

  72. [72]

    Matsumoto, R

    Y. Matsumoto, R. Natsuaki, and A. Hirose, Full-learning rotational quaternion convolutional neural networks and confluence of differently represented data for polsar land classification, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing15, 2914 (2022)

    work page 2022

  73. [73]

    S. Y. Chang, B. Le Saux, S. Vallecorsa, and M. Grossi, Quantum convolutional circuits for earth observation im- age classification, in IGARSS 2022-2022 IEEE Interna- tional Geoscience and Remote Sensing Symposium(IEEE,

    work page 2022

  74. [74]

    Z. N. Aldoski and C. Koren, Impact of traffic sign diver- sityonautonomousvehicles: aliteraturereview,Periodica Polytechnica Transportation Engineering51, 338 (2023)

    work page 2023

  75. [75]

    M. A. Khan, H. Park, and J. Chae, A lightweight convo- lutional neural network (cnn) architecture for traffic sign recognition in urban road networks, Electronics12, 1802 (2023)

    work page 2023

  76. [76]

    S. Y.-C. Chen, T.-C. Wei, C. Zhang, H. Yu, and S. Yoo, Quantum convolutional neural networks for high energy physics data analysis, Phys. Rev. Res.4, 013231 (2022)

    work page 2022

  77. [77]

    S. Y.-C. Chen, T.-C. Wei, C. Zhang, H. Yu, and S. Yoo, Hybrid quantum-classical graph convolutional network, arXiv preprint arXiv:2101.06189 (2021)

    work page arXiv 2021

  78. [78]

    Delgado, K

    A. Delgado, K. E. Hamilton, J.-R. Vlimant, D. Magano, Y. Omar, P. Bargassa, A. Francis, A. Gianelle, L. Sestini, D. Lucchesi,et al., Quantum computing for data analysis in high energy physics, arXiv preprint arXiv:2203.08805 (2022)

    work page arXiv 2022

  79. [79]

    Monaco, O

    S. Monaco, O. Kiss, A. Mandarino, S. Vallecorsa, and M. Grossi, Quantum phase detection generalization from marginal quantum neural network models, Physical Re- view B107, L081105 (2023)

    work page 2023

  80. [80]

    M. Cea, M. Grossi, S. Monaco, E. Rico, L. Tagliacozzo, and S. Vallecorsa, Exploring the phase diagram of the quantum one-dimensional annni model, arXiv preprint arXiv:2402.11022 (2024)

    work page arXiv 2024

Showing first 80 references.