Pith. sign in

REVIEW 3 major objections 4 minor 69 references

Quantum Visual Fields with Neural Amplitude Encoding

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Quantum Visual Field claims that a 5-qubit circuit with a learned Boltzmann amplitude encoder can represent 2D images and 3D signed distance fields more accurately than the prior quantum method QIREN and competitively with classical INR…

desk verdict A novel QINR architecture worth refereeing; the real-valued ansatz expressivity claim is unproven, but the work is honest and the flaws are fixable. read the letter →

arxiv 2508.10900 v2 pith:VBEV222H submitted 2025-08-14 cs.CV

Shuteng Wang , Christian Theobalt , Vladislav Golyanik This is my paper · ORCID
classification cs.CV
keywords quantumimplicitneuralrepresentationamplitudeencodingGibbs–Boltzmannreal-valuedansatzparameterisedcircuitsigneddistancefield3Dshapecompletion
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

The paper introduces Quantum Visual Field (QVF), an implicit neural representation built on a parameterised quantum circuit. QVF encodes a query coordinate (plus an optional latent code) into the amplitudes of an n-qubit state through a learned energy spectrum: a small MLP predicts an energy E, the Gibbs–Boltzmann factor turns it into probabilities, and those probabilities set the squared amplitudes with zero phase. A fully entangled circuit restricted to real-valued unitaries (Pauli-Y rotations plus CNOT entanglers) then evolves the state, and Pauli-Z measurements on the first m qubits produce the image pixel or signed-distance output directly, with no classical decoder. The central claim is that this design beats the previous QINR method QIREN on 2D image reconstruction and is competitive with (and in the reported tables, better than) classical INR baselines such as Siren and DeepSDF-style MLPs for both 2D images and 3D signed distance fields, on a noiseless simulator. If the claim holds, it would show that a quantum circuit can carry the whole implicit representation rather than acting as a Fourier feature extractor for classical layers, and that learned amplitude encoding is a viable way to align classical data with Hilbert-space processing.

What carries the argument

The load-bearing mechanism is the neural amplitude encoder, not the circuit alone. A minimal MLP maps positionally-encoded coordinates and a latent code to an energy spectrum; a Gibbs–Boltzmann factor with inverse temperature β converts that spectrum into probabilities; and those probabilities set the squared amplitudes of the computational-basis input state, with the phase fixed to zero. This is what injects data-dependent non-linearity and a learned frequency structure into the circuit, and it replaces the hand-crafted amplitude embeddings used by earlier quantum auto-encoders. Around it sits the fully entangled real-valued ansatz S(θ): alternating Pauli-Y single-qubit rotations and CNOT entanglers, initialised either as identity blocks or with Gaussian noise scaled by depth, trained by the parameter-shift rule. Local Pauli-Z measurements on the first m qubits extract the output. The paper's trainability argument is the claim that restricting unitaries to the real subspace and using local observables keeps gradient variance from decaying exponentially, which is exactly the role the ansatz plays in the architecture.

What would settle it

Run the QVF ansatz with 20 or more qubits, keeping depth per qubit and the same Boltzmann encoder, and plot gradient variance over random parameters: if the variance decays exponentially in qubit count as in the strongly entangled ansatz, the barren-plateau-avoidance claim fails. Then, on the same 50-image benchmark, fix the circuit and replace the learned energy encoding with a random amplitude encoding of equal dimension: if PSNR stays at the reported level, the learned energy manifold is not the source of the gains. Finally, on near-term hardware with finite shots, measure PSNR versus shot count; the claim that QVF is competitive with classical INRs should be re-evaluated at the shot budget a real device can afford.

Watch

Extended reading notes

Core claim

On its own terms, QVF's discovery is that a QINR can produce visual fields directly from projective measurement once two design choices are made together. First, classical data is embedded by learning an energy manifold: the conditional energy spectrum E(Θ,z) inferred by a tiny MLP is converted by the Gibbs distribution P = exp(−βE)/Z into probabilities that become the squared moduli of the input statevector amplitudes (phases set to zero). Second, the ansatz is restricted to the real Hilbert subspace, built from alternating Pauli-Y rotations and CNOT entanglers, so that complex phases never enter the evolution; the paper argues this 'bounded Haar randomness' keeps the optimisation landscape free of barren plateaus while preserving everything measurable with Pauli-Z observables. Measured expectation values are mapped injectively to signal dimensions. The reported consequence is that with n=5 qubits, depth J=5 and a 128-unit MLP, QVF's best variant reaches 0.54×10^−3 MSE and 32.67 dB PSNR on 50 CIFAR-10 images versus 0.78×10^−3 and 31.03 dB for QIREN, and on three ShapeNet SDFs it reaches 0.27×10^−3 MAE versus 0.48×10^−3 for MLP+Siren and 1.43×10^−3 for the MLP+ReLU baseline. Since the ansatz adds only about 170 parameters, the paper attributes the gains to the encoding plus ansatz inductive bias, not to raw parameter count.

Load-bearing premise

The central bet is that a circuit built only from real-valued rotations and CNOT entanglers loses no measurement-relevant expressivity while keeping gradients from vanishing; the paper supports this with gradient-variance measurements on a simulator, not with a proof, so it is the premise that would most quickly sink the approach at scale.

Editorial extensions

If this is right

  • QVF supports representation of image collections and 3D shape collections through a single latent-conditioned circuit, a capability previous QINR methods did not offer.
  • With identical quantum resources (5 qubits, same simulator), QVF's best variant reports roughly 30% lower MSE and 1.6 dB higher PSNR than QIREN on 50 images.
  • On 3D signed distance fields, QVF's MAE of 0.27×10^−3 is below the classical Siren baseline (0.48×10^−3) and far below the ReLU/DeepSDF baseline (1.43×10^−3), so the ansatz improves rather than degrades representation quality.
  • Finite-shot measurements degrade image quality predictably, and gate-noise simulations show PSNR stays around 25–28 dB at σ=0.1, so near-term hardware use becomes a shot-fidelity trade-off rather than a conceptual obstacle.
  • Parameter counts scale as O(p^2) for the classical encoder and O(nJ) for the ansatz, meaning the quantum part is a low-cost inductive-bias addition to an otherwise classical representation.

Reading between the lines

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

  • If the real-valued ansatz claim survives larger qubit counts, it points to architecture-level barren-plateau avoidance that could be combined with smarter initialisation and error-mitigation, extending beyond QINRs to variational quantum algorithms generally.
  • A natural ablation the paper does not run: replace the learned Boltzmann encoding with a fixed random amplitude encoding of the same dimension; if performance drops sharply, the learned energy manifold—not the circuit—is doing most of the work.
  • The shot-noise curves imply the simulator numbers are effectively the infinite-shot limit; a meaningful comparison on real hardware would fix a wall-clock shot budget and measure PSNR per unit time.
  • Because the energy encoder outputs a frequency spectrum in the Fourier sense of variational quantum models, the QVF construction could be lifted to scene reconstruction or neural rendering tasks where spectral bias matters.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes Quantum Visual Fields (QVF), a quantum implicit neural representation for 2D images and 3D signed distance fields. QVF encodes query coordinates into quantum statevector amplitudes through a learned energy-based Boltzmann distribution, processes them with a fully entangled parameterized circuit built from Pauli-Y rotations and CNOT gates, and extracts the represented field directly from local Pauli-Z measurements without a classical decoder. The method is evaluated on a noiseless PennyLane simulator with n=5 qubits and J=5 layers, compared against the QINR method QIREN on equal qubit budgets and against classical INR baselines (MLP+ReLU, DeepSDF, SIREN) on image and shape collections. The authors report that QVF outperforms QIREN and is competitive with or better than the classical baselines in MSE, PSNR, and 3D MAE, and they demonstrate image inpainting, shape completion, and latent-space interpolation as applications.

Significance. If the reported results hold, QVF is a meaningful advance for QINRs: it is the first QINR in this line to support collections of images and 3D SDFs, it avoids the classical decoder used by QIREN, and it includes an ablation of the ansatz depth, width, and latent dimension. The empirical comparisons are run with multiple repetitions and reported with means and standard deviations, and the equal-qubit-budget comparison with QIREN is a fair and useful baseline. The significance is conditional, however, on two issues: the theoretical claim that the real-valued ansatz retains full expressivity and good gradient behavior at larger scales is not established, and key experimental hyperparameters (notably beta and the measurement-shot count) are not reported. These do not invalidate the n=5 simulator findings, but they do limit the paper's broader claim that QVF is a scalable QINR architecture.

major comments (3)
  1. [Section 4.2; Appendix D.3] The claim in Section 4.2 that restricting the ansatz to real-valued Pauli-Y rotations and CNOT gates 'maintains full expressivity while significantly simplifying the optimisation landscape' is load-bearing for the scalability of QVF, but the evidence provided is insufficient. Appendix D.3's Lemma 1 invokes the Fourier framework of Schuld et al. for circuits of the form U(x)=W2 g(x) W1, whereas QVF's input state is prepared by amplitude encoding with non-negative real amplitudes derived from a Boltzmann distribution; the lemma does not directly transfer to this setting. Figure 5 reports gradient variance over a small range of qubit counts and without an asymptotic fit, so it does not establish that the real-valued restriction avoids barren plateaus at larger n. Please either provide a proof of the expressivity and variance-scaling claims, or explicitly restrict the claims to the tested n=5 regime.
  2. [Section 5.2 and Algorithm 1] Two central experimental hyperparameters are never reported. The inverse temperature beta governing the Boltzmann encoding in Eq. (2) is listed as an input in Algorithm 1, but no value is given in the main text or appendix. Similarly, the number of measurement shots used for the quantitative results in Tables 2--4 is not stated; shot counts (100 to 10^4) appear only in Appendix E for the qualitative figure. If the simulator uses analytic expectation values rather than finite-shot sampling, that should be stated explicitly; otherwise the finite-shot statistics should be reported. Without these values the central empirical comparisons are not reproducible, and the effect of measurement noise on the reported differences cannot be assessed.
  3. [Section 4.3, Eq. (11)] The paper's headline distinction over QIREN is that QVF 'does not rely on classical post-processing'. This is accurate for the readout stage, since circuit outputs are obtained by direct measurements. However, the application protocol in Eq. (11) performs a classical MAP optimization of a latent code z after training, and the energy encoder is a classical MLP before the circuit. The scope of the no-post-processing claim should be stated precisely, because as written it overstates the purity of the quantum pipeline and invites a misleading comparison with QIREN's classical decoder.
minor comments (4)
  1. [Eq. (3)] Equation (3) is typeset incorrectly: the integral sign and the differential dTheta are misplaced, and the intended expression Z = integral exp(-beta E(Theta,z)) dTheta should be written clearly.
  2. [Eq. (10)] The summation notation in Eq. (10), written as sum_{i,j=1}^{W,M}, is unconventional; please use explicit double sums or a standard multi-index notation.
  3. [Table 3] Table 3 labels one baseline as 'DeepSDF [41]', while the text describes it as an MLP with ReLU activations; please clarify the exact architecture, latent dimension, and parameter count used for this baseline so that the comparison is unambiguous.
  4. [Section 6] The statement that QVF 'trains faster than all compared methods' is not supported by wall-clock or per-epoch runtime measurements; Figure 6 shows learning curves, which measure convergence behaviour, not training speed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; QVF's accuracy claims are benchmarked externally against QIREN and classical INRs, and no fitted parameter is renamed as a prediction.

full rationale

QVF's derivation chain is self-contained in the relevant sense. The learnable energy module f(γ(Θ), z) outputs energies E, which are converted by the Gibbs formula P = exp(−βE)/Z and square-rooted into amplitudes α_i. These amplitudes are then evolved by the fixed-structure ansatz Ŝ(θ) and read out via Pauli-Z expectation values. Every component is trained end-to-end on the reconstruction loss in Eq. (10) against ground-truth pixel or SDF values; no target quantity is used to define a parameter, and no fitted value is renamed as a prediction. The central comparison—QVF versus QIREN and versus SIREN/DeepSDF—is against external baselines on held-out metrics (Tables 2 and 3), so the accuracy claims are not forced by construction. The only self-citation is the authors' own 3D-QAE [43], which is described as inspiration and used as a baseline and hand-crafted encoding comparison; it is not invoked to justify QVF's accuracy, trainability, or expressivity. Section 4.2's assertion that the real-valued Ry+CNOT ansatz 'maintains full expressivity while significantly simplifying the optimisation landscape' is unproven and could be a correctness risk at larger qubit counts, but the claim is an assumption about the ansatz, not a step that reduces to its own inputs. Fig. 5 and Appendix D.3 provide supporting evidence but do not make the central result definitionally true. Accordingly, no circular step can be exhibited with a specific equation-to-equation reduction.

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

The central result is empirical; it rests on standard quantum mechanics (Born rule, unitary evolution), standard theorems (Solovay-Kitaev, CLT), and the design assumption that real-valued circuits preserve expressivity for Z-basis measurements. One free hyperparameter, beta, is not reported.

free parameters (1)
  • beta (inverse temperature) = not reported (default likely 1)
    Appears in Eq. (2) in the Boltzmann encoding P = exp(-beta E)/Z; controls the sharpness of the amplitude distribution and is never specified in the experimental section.
assumptions (4)
  • domain assumption Real-valued unitaries composed of Pauli-Y rotations and CNOT entanglers are universal for representing Z-basis measurement probabilities and preserve expressivity.
    Sec. 4.2 states the design 'maintains full expressivity while significantly simplifying the optimisation landscape' but no proof is given; this is load-bearing because the architecture's performance depends on it.
  • standard math Quantum circuits of sufficient depth can approximate arbitrary unitaries (Solovay-Kitaev), so depth J is a valid capacity knob.
    Cited in Sec. 4.2 and App. A.
  • standard math Central Limit Theorem: the measurement statistics of the circuit output approximate a Gaussian, justifying the likelihood in Eq. (9).
    Stated in Sec. 4.3.
  • domain assumption The Fourier structure of parameterized quantum circuits (from Schuld et al. [50]) applies to the encoding used here, giving the model frequency expressivity.
    Used in App. D.3 Lemma 1 to argue the energy encoding controls the frequency spectrum.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Visual Fields with Neural Amplitude Encoding." pith.science (2026). https://pith.science/paper/VBEV222H

@misc{pith2026250810900,
  author       = {Pith},
  title        = {Pith review of: Quantum Visual Fields with Neural Amplitude Encoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VBEV222H}},
  note         = {Machine review of arXiv:2508.10900}
}
read the original abstract

Quantum Implicit Neural Representations (QINRs) have emerged as a promising paradigm that leverages parametrised quantum circuits to encode and process classical information. However, significant challenges remain in areas such as ansatz architecture design, the effective utility of quantum-mechanical properties, training efficiency, and the integration with classical modules. This paper advances the field by introducing a novel QINR architecture for 2D image and 3D geometric field learning, which we collectively refer to as Quantum Visual Field (QVF). QVF encodes classical data into quantum statevectors using neural amplitude encoding grounded in a learnable energy manifold, ensuring meaningful Hilbert-space embeddings. Our ansatz follows a fully entangled design of learnable parametrised quantum circuits, with quantum (unitary) operations performed in the real Hilbert space, resulting in numerically stable training with fast convergence. QVF does not rely on classical post-processing -- in contrast to the previous QINR learning approach -- and directly employs measurements to extract learned signals encoded in the ansatz. Experiments on a quantum hardware simulator demonstrate that QVF outperforms an existing quantum approach and competes with widely used classical foundational baselines in terms of visual representation accuracy across various metrics and model characteristics. We also show applications of QVF in 2D and 3D field completion and 3D shape interpolation, highlighting its practical potential. Project page: https://4dqv.mpi-inf.mpg.de/QVF/.

Figures

Figures reproduced from arXiv: 2508.10900 by the authors.

Figure 1
Figure 1. Our learnable coordinate-based QVF model can represent various visual fields: (a) Schematic diagram of the architecture; (b) 2D image representation of a moderate resolution (400×350 pixels) [19]; (c) Latent space interpolation of 3D signed distance fields [13]. quantum data encoding methods [11, 43, 46, 60]—which (unreasonably) assume that handcrafted embeddings align with the inductive biases of quantum circuits—w… view at source ↗
Figure 2
Figure 2. Overview of the proposed QVF model, a QML framework for visual representation learning. Query coordinates Θ encoded using γ (positional encoding) concatenated with the conditioning latent code z are used to infer the energy spectrum E of a quantum system, associated with Boltzmann-regulated statistical uncertainty P . The inferred statistical property is leveraged in encoding the classical data into quantum statevec… view at source ↗
Figure 3
Figure 3. Representative pure qubit states on the Bloch sphere. Once the classical data is en￾coded into |ψin(Θ, z)⟩, it is pro￾cessed by our learnable PQC or ansatz Sˆ(θ) within a high￾dimensional Hilbert space. Our goal is a compact and expressive PQC for QINR learning. Unre￾stricted transformations of the Hilbert space can require deep quantum gate sequences and can lead to training instabilities such as barren plateaus4 .… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: QVF Initialisation: Schematic circuit module initialised with identity (top) and Gaussian (bottom) schemes. that the composite operation Sˆ(θ)j = SRFR chained over all layers j is equivalent to a zero circuit depth (identity cir￾cuit) before training. Note that while t…
Figure 6
Figure 6. Figure 6: (a) Reconstructed images during training: (top) our QVF [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Visualisation of the reconstructed images (CIFAR-10 [ [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: (a): Geometry representation using QVF. Each shape is [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Ablation study with modules influencing the model [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Bloch sphere visualisa￾tion of qubit states. Qubit 0: |ψ⟩ = √1 2 (|0⟩ + |1⟩), qubit 1: |ψ⟩ = |1⟩. Qubits. The fundamen￾tal information blocks of a quantum process￾ing unit (QPU) are qubits, i.e., the ana￾logues of bits in clas￾sical computing. Un￾like classical bits d…
Figure 11
Figure 11. Figure 11: Visualisation of ansatz designs and their exemplary induced traversable quantum states within Hilbert space: strongly entangled [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 13
Figure 13. Figure 13: Some image inpainting results obtained with QVF pre [PITH_FULL_IMAGE:figures/full_fig_p016_13.png]
Figure 14
Figure 14. Figure 14: Visualisations of 3D shape completion from partial and noisy input depth maps using a pre-trained QVF; [PITH_FULL_IMAGE:figures/full_fig_p017_14.png]
Figure 15
Figure 15. Figure 15: (a) Comparison of geometric representations using QVF and the classical baseline [ [PITH_FULL_IMAGE:figures/full_fig_p017_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

69 extracted references · 62 canonical work pages

  1. [1]

    Quantum error cor- rection below the surface code threshold

    Google Quantum AI and Collaborators. Quantum error cor- rection below the surface code threshold. Nature, 2024. 15

    work page 2024

  2. [2]

    Quantum-classical hybrid machine learning for image classification

    Mahabubul Alam, Satwik Kundu, Rasit Onur Topaloglu, and Swaroop Ghosh. Quantum-classical hybrid machine learning for image classification. In International Conference On Computer Aided Design (ICCAD), 2021. 15

    work page 2021

  3. [3]

    Quantum boltzmann machine

    Mohammad H Amin, Evgeny Andriyash, Jason Rolfe, Bo- hdan Kulchytskyy, and Roger Melko. Quantum boltzmann machine. Physical Review X, 8(2):021050, 2018. 4, 13

    work page 2018

  4. [4]

    3D Scalable Quantum Convolutional Neural Networks for Point Cloud Data Processing in Classification Applications

    Hankyul Baek, Won Joon Yun, and Joongheon Kim. 3d scalable quantum convolutional neural networks for point cloud data processing in classification applications. arXiv preprint 2210.09728, 2022. 3

    work page Pith review arXiv 2022

  5. [5]

    Boltzmann distributions on a quantum computer via active cooling

    Carter Ball and Thomas D Cohen. Boltzmann distributions on a quantum computer via active cooling. Nuclear Physics A, 1038:122708, 2023. 4

    work page 2023

  6. [6]

    Parameterized quantum circuits as machine learn- ing models

    Marcello Benedetti, Erika Lloyd, Stefan Sack, and Mattia Fiorentini. Parameterized quantum circuits as machine learn- ing models. Quantum Science and Technology, 4(4):043001,

    work page

  7. [7]

    Q-match: Iterative shape matching via quantum annealing

    Marcel Seelbach Benkner, Zorah L¨ahner, Vladislav Golyanik, Christof Wunderlich, Christian Theobalt, and Michael Moeller. Q-match: Iterative shape matching via quantum annealing. In International Conference on Computer Vision (ICCV), pages 7586–7596, 2021. 2

    work page 2021

  8. [8]

    Penny- lane: Automatic differentiation of hybrid quantum-classical computations

    Ville Bergholm, Josh Izaac, Maria Schuld, Christian Gogolin, Shahnawaz Ahmed, Vishnu Ajith, M Sohaib Alam, Guillermo Alonso-Linaje, B AkashNarayanan, Ali Asadi, et al. Penny- lane: Automatic differentiation of hybrid quantum-classical computations. arXiv preprint 1811.04968, 2018. 2, 6, 7, 14, 16

    arXiv 2018

Show all 69 references
  1. [9]

    Quantum machine learning

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Reben- trost, Nathan Wiebe, and Seth Lloyd. Quantum machine learning. Nature, 549(7671), 2017. 3

    2017

  2. [10]

    Quantum permutation synchronization

    Tolga Birdal, Vladislav Golyanik, Christian Theobalt, and Leonidas J Guibas. Quantum permutation synchronization. In Computer Vision and Pattern Recognition (CVPR), 2021. 2

    2021

  3. [11]

    Quantum autoen- coders to denoise quantum data

    Dmytro Bondarenko and Polina Feldmann. Quantum autoen- coders to denoise quantum data. Physical review letters, 124 (13):130502, 2020. 2

    2020

  4. [12]

    Cost function dependent barren plateaus in shallow parametrized quantum circuits

    Marco Cerezo, Akira Sone, Tyler V olkoff, Lukasz Cincio, and Patrick J Coles. Cost function dependent barren plateaus in shallow parametrized quantum circuits. Nature communica- tions, 12(1):1791, 2021. 5, 14

    2021

  5. [13]

    Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu

    Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. ShapeNet: An Information-Rich 3D Model Repos- itory. arXiv preprint 1512.03012, 2015. 2, 6, 8

    2015 arXiv

  6. [14]

    Stylebank: An explicit representation for neural image style transfer

    Dongdong Chen, Lu Yuan, Jing Liao, Nenghai Yu, and Gang Hua. Stylebank: An explicit representation for neural image style transfer. In Computer Vision and Pattern Recognition (CVPR), pages 1897–1906, 2017. 2

    1906

  7. [15]

    Learning contin- uous image representation with local implicit image function

    Yinbo Chen, Sifei Liu, and Xiaolong Wang. Learning contin- uous image representation with local implicit image function. In Computer Vision and Pattern Recognition (CVPR), 2021. 2

    2021

  8. [16]

    Quantum convolutional neural networks

    Iris Cong, Soonwon Choi, and Mikhail D Lukin. Quantum convolutional neural networks. Nature Physics, 15(12):1273– 1278, 2019. 3

    2019

  9. [17]

    Coin: Compression with implicit neural representations

    Emilien Dupont, Adam Goli´nski, Milad Alizadeh, Yee Whye Teh, and Arnaud Doucet. Coin: Compression with implicit neural representations. arXiv preprint 2103.03123, 2021. 2

    2021 arXiv

  10. [18]

    Quantum multi- model fitting

    Matteo Farina, Luca Magri, Willi Menapace, Elisa Ricci, Vladislav Golyanik, and Federica Arrigoni. Quantum multi- model fitting. In Computer Vision and Pattern Recognition (CVPR), pages 13640–13649, 2023. 2

    2023

  11. [19]

    Jonathan P. Gardner. James Webb Space Tele- scope. https : / / webb . nasa . gov / content / multimedia/images.html, 2022. 2, 6, 7

    2022

  12. [20]

    A quantum com- putational approach to correspondence problems on point sets

    Vladislav Golyanik and Christian Theobalt. A quantum com- putational approach to correspondence problems on point sets. In Computer Vision and Pattern Recognition (CVPR), 2020. 2

    2020

  13. [21]

    An initialization strategy for addressing barren plateaus in parametrized quantum circuits

    Edward Grant, Leonard Wossnig, Mateusz Ostaszewski, and Marcello Benedetti. An initialization strategy for addressing barren plateaus in parametrized quantum circuits. Quantum, 3:214, 2019. 5

    2019

  14. [22]

    Der massbegriff in der theorie der kontinuier- lichen gruppen

    Alfred Haar. Der massbegriff in der theorie der kontinuier- lichen gruppen. Annals of Mathematics, 34(1):147–169, 1933. 2

    1933

  15. [23]

    Hybrid quantum network for classification of finance and mnist data

    Gerhard Hellstem. Hybrid quantum network for classification of finance and mnist data. In International Conference on Software Architecture Companion (ICSA-C), 2021. 15

    2021

  16. [24]

    Connecting ansatz expressibility to gradient magni- tudes and barren plateaus

    Zo¨e Holmes, Kunal Sharma, Marco Cerezo, and Patrick J Coles. Connecting ansatz expressibility to gradient magni- tudes and barren plateaus. PRX Quantum, 3(1):010313, 2022. 14

    2022

  17. [25]

    Hands-on bayesian neural networks—a tutorial for deep learning users

    Laurent Valentin Jospin, Hamid Laga, Farid Boussaid, Wray Buntine, and Mohammed Bennamoun. Hands-on bayesian neural networks—a tutorial for deep learning users. IEEE Computational Intelligence Magazine, 17(2):29–48, 2022. 14

    2022

  18. [26]

    Design space explo- ration of hybrid quantum–classical neural networks

    Muhammad Kashif and Saif Al-Kuwari. Design space explo- ration of hybrid quantum–classical neural networks. Electron- ics, 10(23):2980, 2021. 15

    2021

  19. [27]

    q-means: A quantum algorithm for unsu- pervised machine learning

    Iordanis Kerenidis, Jonas Landman, Alessandro Luongo, and Anupam Prakash. q-means: A quantum algorithm for unsu- pervised machine learning. Advances in Neural Information Processing Systems (NeurIPS), 32, 2019. 13

    2019

  20. [28]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint 1412.6980, 2014. 6

    2014 arXiv

  21. [29]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. University of Toronto,

  22. [30]

    3d neural scene representations for visuomotor control

    Yunzhu Li, Shuang Li, Vincent Sitzmann, Pulkit Agrawal, and Antonio Torralba. 3d neural scene representations for visuomotor control. In Conference on Robot Learning, pages 112–123. PMLR, 2022. 2

    2022

  23. [31]

    Hybrid quantum-classical convo- lutional neural networks

    Junhua Liu, Kwan Hui Lim, Kristin L Wood, Wei Huang, Chu Guo, and He-Liang Huang. Hybrid quantum-classical convo- lutional neural networks. Science China Physics, Mechanics & Astronomy, 64(9):290311, 2021. 15 9

    2021

  24. [32]

    Quan- tum principal component analysis

    Seth Lloyd, Masoud Mohseni, and Patrick Rebentrost. Quan- tum principal component analysis. Nature Physics, 10(9): 631–633, 2014. 13

    2014

  25. [33]

    Marching cubes: A high resolution 3d surface construction algorithm

    William E Lorensen and Harvey E Cline. Marching cubes: A high resolution 3d surface construction algorithm. In Seminal graphics: pioneering efforts that shaped the field, pages 347–

  26. [34]

    Barren plateaus in quan- tum neural network training landscapes

    Jarrod R McClean, Sergio Boixo, Vadim N Smelyanskiy, Ryan Babbush, and Hartmut Neven. Barren plateaus in quan- tum neural network training landscapes. Nature communica- tions, 9(1):4812, 2018. 14

    2018

  27. [35]

    QuCOOP: A versatile frame- work for solving composite and binary-parametrised prob- lems on quantum annealers

    Natacha Kuete Meli, Vladislav Golyanik, Marcel Seelbach Benkner, and Michael Moeller. QuCOOP: A versatile frame- work for solving composite and binary-parametrised prob- lems on quantum annealers. In Computer Vision and Pattern Recognition (CVPR), 2025. 2

    2025

  28. [36]

    Quantum state preparation using tensor networks

    Ar Melnikov, Alena Termanova, Sergey Dolgov, Florian Neukart, and MR Perelshtein. Quantum state preparation using tensor networks. Quantum Science and Technology, 8 (3):035027, 2023. 8

    2023

  29. [37]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. Communications of the ACM , 65(1):99–106, 2021. 4

    2021

  30. [38]

    Quantum circuit learning

    Kosuke Mitarai, Makoto Negoro, Masahiro Kitagawa, and Keisuke Fujii. Quantum circuit learning. Physical Review A, 98(3):032309, 2018. 6

    2018

  31. [39]

    Implicit neural representation in medical imaging: A comparative survey

    Amirali Molaei, Amirhossein Aminimehr, Armin Tavakoli, Amirhossein Kazerouni, Bobby Azad, Reza Azad, and Dorit Merhof. Implicit neural representation in medical imaging: A comparative survey. InInternational Conference on Computer Vision (ICCV), pages 2381–2391, 2023. 2

    2023

  32. [40]

    Nielsen and Isaac L

    Michael A. Nielsen and Isaac L. Chuang. Quantum Compu- tation and Quantum Information. Massachusetts Institute of Technology, 2000. 3

    2000

  33. [41]

    Deepsdf: Learning continuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard New- combe, and Steven Lovegrove. Deepsdf: Learning continuous signed distance functions for shape representation. In Com- puter Vision and Pattern Recognition (CVPR), 2019. 2, 6, 8, 17

    2019

  34. [42]

    On the spectral bias of neural networks

    Nasim Rahaman, Aristide Baratin, Devansh Arpit, Felix Draxler, Min Lin, Fred Hamprecht, Yoshua Bengio, and Aaron Courville. On the spectral bias of neural networks. In International conference on machine learning, pages 5301–

  35. [43]

    3D-QAE: Fully quantum auto-encoding of 3d point clouds

    Lakshika Rathi, Edith Tretschk, Christian Theobalt, Rishabh Dabral, and Vladislav Golyanik. 3D-QAE: Fully quantum auto-encoding of 3d point clouds. In The British Machine Vision Conference (BMVC), 2023. 2, 3, 4, 7, 16

    2023

  36. [44]

    Quan- tum support vector machine for big data classification

    Patrick Rebentrost, Masoud Mohseni, and Seth Lloyd. Quan- tum support vector machine for big data classification. Physi- cal review letters, 113(13):130503, 2014. 13

    2014

  37. [45]

    Kilonerf: Speeding up neural radiance fields with thousands of tiny mlps

    Christian Reiser, Songyou Peng, Yiyi Liao, and Andreas Geiger. Kilonerf: Speeding up neural radiance fields with thousands of tiny mlps. In International Conference on Com- puter Vision (ICCV), 2021. 8

    2021

  38. [46]

    On the importance of data encoding in quantum boltzmann methods

    Merel A Schalkers and Matthias M¨oller. On the importance of data encoding in quantum boltzmann methods. Quantum Information Processing, 23(1):20, 2024. 2

    2024

  39. [47]

    Quantum deep equilibrium models

    Philipp Schleich, Marta Skreta, Lasse Kristensen, Rodrigo Vargas-Hernandez, and Al´an Aspuru-Guzik. Quantum deep equilibrium models. In Advances in Neural Information Pro- cessing Systems (NeurIPS), pages 31940–31967, 2024. 4

    2024

  40. [48]

    An introduction to quantum machine learning

    Maria Schuld, Ilya Sinayskiy, and Francesco Petruccione. An introduction to quantum machine learning. Contemporary Physics, 56(2):172–185, 2015. 13

    2015

  41. [49]

    Circuit-centric quantum classifiers

    Maria Schuld, Alex Bocharov, Krysta M Svore, and Nathan Wiebe. Circuit-centric quantum classifiers. Physical Review A, 101(3):032308, 2020. 15

    2020

  42. [50]

    Ef- fect of data encoding on the expressive power of variational quantum-machine-learning models

    Maria Schuld, Ryan Sweke, and Johannes Jakob Meyer. Ef- fect of data encoding on the expressive power of variational quantum-machine-learning models. Physical Review A, 103 (3):032430, 2021. 1, 13, 15

    2021

  43. [51]

    Quantum machine learning for image classification

    Arsenii Senokosov, Alexandr Sedykh, Asel Sagingalieva, Basil Kyriacou, and Alexey Melnikov. Quantum machine learning for image classification. Machine Learning: Science and Technology, 5(1):015040, 2024. 15

    2024

  44. [52]

    Convolution fil- ter embedded quantum gate autoencoder

    Kodai Shiba, Katsuyoshi Sakamoto, Koichi Yamaguchi, Di- nesh Bahadur Malla, and Tomah Sogabe. Convolution fil- ter embedded quantum gate autoencoder. arXiv preprint 1906.01196, 2019. 3

    1906 arXiv

  45. [53]

    Implicit neural represen- tations with periodic activation functions

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural represen- tations with periodic activation functions. In Advances in Neural Information Processing Systems (NeurIPS) , pages 7462–7473, 2020. 4, 8

    2020

  46. [54]

    Neural geometric level of detail: Real-time rendering with implicit 3d shapes

    Towaki Takikawa, Joey Litalien, Kangxue Yin, Karsten Kreis, Charles Loop, Derek Nowrouzezahrai, Alec Jacobson, Mor- gan McGuire, and Sanja Fidler. Neural geometric level of detail: Real-time rendering with implicit 3d shapes. In Com- puter Vision and Pattern Recognition (CVPR)...

    2021

  47. [55]

    Advances in neural rendering

    Ayush Tewari, Justus Thies, Ben Mildenhall, Pratul Srini- vasan, Edgar Tretschk, Wang Yifan, Christoph Lassner, Vin- cent Sitzmann, Ricardo Martin-Brualla, Stephen Lombardi, et al. Advances in neural rendering. In Computer Graphics Forum, pages 703–735. Wiley Online Library, 2022. 8

    2022

  48. [56]

    Subtleties in the trainability of quantum machine learning models

    Supanut Thanasilp, Samson Wang, Nhat Anh Nghiem, Patrick Coles, and Marco Cerezo. Subtleties in the trainability of quantum machine learning models. Quantum Machine Intel- ligence, 5(1):21, 2023. 5, 14

    2023

  49. [57]

    Non- rigid neural radiance fields: Reconstruction and novel view synthesis of a dynamic scene from monocular video

    Edgar Tretschk, Ayush Tewari, Vladislav Golyanik, Michael Zollh¨ofer, Christoph Lassner, and Christian Theobalt. Non- rigid neural radiance fields: Reconstruction and novel view synthesis of a dynamic scene from monocular video. In International Conference on Computer Vision (...

    2021

  50. [58]

    Neural feature fusion fields: 3d distillation of self- supervised 2d image representations

    Vadim Tschernezki, Iro Laina, Diane Larlus, and Andrea Vedaldi. Neural feature fusion fields: 3d distillation of self- supervised 2d image representations. In International Con- ference on 3D Vision (3DV), pages 443–453, 2022. 2

    2022

  51. [59]

    Noise- induced barren plateaus in variational quantum algorithms

    Samson Wang, Enrico Fontana, Marco Cerezo, Kunal Sharma, Akira Sone, Lukasz Cincio, and Patrick J Coles. Noise- induced barren plateaus in variational quantum algorithms. Nature communications, 12(1):6961, 2021. 14 10

    2021

  52. [60]

    Data encoding patterns for quantum computing

    Manuela Weigold, Johanna Barzen, Frank Leymann, and Marie Salm. Data encoding patterns for quantum computing. In Conference on Pattern Languages of Programs, 2020. 2

    2020

  53. [61]

    General parameter-shift rules for quantum gradients

    David Wierichs, Josh Izaac, Cody Wang, and Cedric Yen-Yu Lin. General parameter-shift rules for quantum gradients. Quantum, 6:677, 2022. 13

    2022

  54. [62]

    Neural fields in visual computing and beyond

    Yiheng Xie, Towaki Takikawa, Shunsuke Saito, Or Litany, Shiqin Yan, Numair Khan, Federico Tombari, James Tompkin, Vincent Sitzmann, and Srinath Sridhar. Neural fields in visual computing and beyond. In Computer Graphics Forum, pages 641–676. Wiley Online Library, 2022. 1

    2022

  55. [63]

    Adiabatic quantum comput- ing for multi object tracking

    Jan-Nico Zaech, Alexander Liniger, Martin Danelljan, Dengxin Dai, and Luc Van Gool. Adiabatic quantum comput- ing for multi object tracking. In Computer Vision and Pattern Recognition (CVPR), pages 8811–8822, 2022. 2

    2022

  56. [64]

    Escaping from the barren plateau via gaussian initializations in deep variational quantum circuits

    Kaining Zhang, Liu Liu, Min-Hsiu Hsieh, and Dacheng Tao. Escaping from the barren plateau via gaussian initializations in deep variational quantum circuits. Advances in Neural Information Processing Systems, 35:18612–18627, 2022. 5

    2022

  57. [65]

    Quan- tum implicit neural representations

    Jiaming Zhao, Wenbo Qiao, Peng Zhang, and Hui Gao. Quan- tum implicit neural representations. In International Confer- ence on Machine Learning (ICML), 2024. 1, 3, 6, 7, 8, 14, 15, 16 11 Quantum Visual Fields with Neural Amplitude Encoding —–Appendix—– This appendix supplement...

    2024

  58. [66]

    −i sin(θ 2) −i sin(θ

  59. [67]

    (18) The Pauli operators ˆX, ˆY, ˆZ represent specific instances of the above rotation operators, i.e

    cos( θ 2) , ˆRy(τ) = cos(τ 2 ) − sin(τ 2 ) sin(τ 2 ) cos( τ 2 ) , and ˆRz(γ) = e−iγ 2 0 0 eiγ 2 . (18) The Pauli operators ˆX, ˆY, ˆZ represent specific instances of the above rotation operators, i.e. rotations by π radians along thex-,y- andz-axes, respectively. These operato...

  60. [68]

    PQC have been shown to be asymptotic universal function approximators [6, 50]

    a variational ansatz, which performs quantum transfor- mation on the quantum states. PQC have been shown to be asymptotic universal function approximators [6, 50]. Several standardised QML algorithms have been explored, including quantum principal component analysis [32], quan...

  61. [69]

    as in the main matter (Sec. 5.2). We report the results with different levels of quantum gate perturbation ratioσ in Table 4 to quantify the degradation in performance under various noise regimes. As expected, increasingσ leads to a decrease in 2D image PSNR. Even withσ = 0.1,...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.