pith. sign in

arxiv: 2604.16662 · v1 · submitted 2026-04-17 · 🪐 quant-ph · eess.IV

Resource-Efficient Quantum-Enhanced Compressive Imaging via Quantum Classical co-Design

Haowei Shi , Visuttha Manthamkarn , Christopher M. Jones , Zheshen Zhang , Quntao Zhuang This is my paper

Pith reviewed 2026-05-10 08:09 UTC · model grok-4.3

classification 🪐 quant-ph eess.IV
keywords quantum imagingcompressive sensingprincipal component analysisquantum squeezingresource efficiencyquantum-classical co-designimage reconstructionimage classification
0
0 comments X

The pith

Co-design between quantum squeezing and classical PCA reduces the number of squeezed modes needed for high-performance imaging.

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

This paper demonstrates that quantum-enhanced imaging does not require squeezing every pixel or spatial mode independently. Instead, classical principal component analysis applied to training data identifies a small set of the most informative modes, and squeezing is applied only to those. The resulting joint quantum-classical framework achieves high-accuracy image classification and high-fidelity reconstruction while using substantially fewer squeezed modes than conventional pixel-wise schemes. Because quantum resources such as squeezing strength and number of modes are costly to generate and maintain, this selective allocation makes quantum imaging more scalable with image size. The approach treats quantum hardware and classical post-processing as a single co-designed system rather than separate layers.

Core claim

Integrating quantum resource allocation with the guidance from classical compressive imaging, via co-design between the quantum hardware layer and the classical software layer, substantially reduces the required quantum resources. Principal component analysis identifies a low-dimensional principal component subspace for measurement, and squeezing is applied selectively to the most informative spatial modes corresponding to these principal components. Numerical experiments show that high-accuracy image classification and high-fidelity image reconstruction can be achieved with significantly fewer squeezed modes compared to pixel-wise squeezing.

What carries the argument

PCA-guided selective squeezing of principal-component spatial modes, which concentrates quantum noise reduction on the low-dimensional subspace identified by classical analysis instead of applying it uniformly across all pixels.

If this is right

  • High-accuracy image classification remains possible when squeezing is limited to a small number of principal-component modes rather than every pixel.
  • High-fidelity image reconstruction is preserved despite the reduction in the number of squeezed modes.
  • Quantum resource requirements no longer scale linearly with image dimension but instead with the dimension of the principal-component subspace.
  • The same co-design principle applies to both classification and reconstruction tasks within the compressive imaging pipeline.

Where Pith is reading between the lines

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

  • The same PCA-guided selection could be used in other quantum sensing tasks where classical data provides a reliable prior on which modes carry the signal.
  • Adaptive hardware implementations could retrain the PCA subspace on incoming data to reallocate squeezing resources in real time.
  • Combining this selective squeezing with additional classical compression steps beyond PCA may yield further resource savings in practical systems.
  • Experimental validation on optical setups with limited squeezing sources would test whether the numerical gains survive realistic loss and detector noise.

Load-bearing premise

The principal components extracted by classical PCA on training data remain the most informative modes to squeeze once quantum noise reduction is selectively applied, and the numerical results hold beyond the specific image datasets and noise models tested.

What would settle it

A side-by-side quantum imaging experiment that applies the same total squeezing resource either to all pixels or only to the top PCA modes and measures whether classification accuracy or reconstruction error matches the claimed improvement.

Figures

Figures reproduced from arXiv: 2604.16662 by Christopher M. Jones, Haowei Shi, Quntao Zhuang, Visuttha Manthamkarn, Zheshen Zhang.

Figure 1
Figure 1. Figure 1: The first 10 principal components of MNIST and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mixing a mode in squeezed state with modes in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Schematic of conventional pixel-wise squeezing-enhanced RF-photonics with a pixel-wise homodyne detector array. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic of the PCA imaging. EOM: electro-optic modulator. LO: local oscillator. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) Accuracy of the MNIST data classification with [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Accuracy of the CIFAR-10 data classification with [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Imaging performance of the 50-PCA with convolutional neural network data processing. Signal photon number per [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
read the original abstract

Quantum sensing can enhance imaging performance by reducing measurement noise below the classical limit, thereby improving the signal-to-noise ratio (SNR) of acquired data. In conventional quantum imaging schemes, squeezing is applied independently to each pixel or spatial mode, leading to a quantum resource cost that scales linearly with image dimension. This approach implicitly separates quantum enhancement from classical post-processing, treating them as independent layers. In this work, we demonstrate that integrating quantum resource allocation with the guidance from classical compressive imaging, via co-design between the quantum hardware layer and the classical software layer, substantially reduces the required quantum resources. We employ principal component analysis (PCA) to identify a low-dimensional principal component subspace for measurement and apply squeezing selectively to the most informative spatial modes corresponding to these principal components. Our numerical experiments show that high-accuracy image classification and high-fidelity image reconstruction can be achieved with significantly fewer squeezed modes compared to pixel-wise squeezing. Our results establish a joint quantum classical co-design framework for resource-efficient quantum-enhanced imaging.

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 / 1 minor

Summary. The manuscript claims that integrating quantum resource allocation with classical compressive imaging via co-design—specifically, using PCA on training data to identify a low-dimensional subspace and applying squeezing selectively only to the corresponding spatial modes—substantially reduces the number of required squeezed modes compared to pixel-wise squeezing, while still enabling high-accuracy image classification and high-fidelity reconstruction.

Significance. If the central claim holds after addressing the noted issues, the work offers a concrete route to lowering quantum resource overhead in imaging by leveraging classical subspace selection, which could make quantum-enhanced sensing more scalable. The co-design idea itself is a positive contribution to bridging quantum hardware and classical post-processing.

major comments (2)
  1. [Numerical Experiments] The numerical experiments section provides no details on the image datasets, noise models, error bars, or exact squeezing parameters used to support the performance claims. Without these, the reported accuracy and resource savings cannot be independently assessed or reproduced.
  2. [Proposed Method] The method computes PCA once on classical training data and then applies selective squeezing to those modes without re-deriving or validating the basis under the modified noise covariance induced by squeezing. Because squeezing changes the effective signal-to-noise structure, the classically dominant modes may no longer be optimal; the manuscript must either iterate the basis selection under the quantum noise model or demonstrate invariance of the subspace to support the resource-efficiency claim.
minor comments (1)
  1. [Abstract] The abstract states that 'significantly fewer squeezed modes' suffice but does not report quantitative reduction factors or direct comparisons (e.g., number of modes vs. accuracy curves), which would clarify the practical gains.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and for recognizing the potential of our quantum-classical co-design framework. We address each major comment below and have revised the manuscript to strengthen the presentation and analysis.

read point-by-point responses

  1. Referee: [Numerical Experiments] The numerical experiments section provides no details on the image datasets, noise models, error bars, or exact squeezing parameters used to support the performance claims. Without these, the reported accuracy and resource savings cannot be independently assessed or reproduced.

    Authors: We agree that additional details are required for reproducibility. In the revised manuscript we have inserted a new subsection titled 'Numerical Setup and Parameters' that specifies: the datasets employed (MNIST for classification and a collection of 28x28 grayscale natural images for reconstruction), the complete noise model (vacuum fluctuations, 3-9 dB squeezing, and additive classical readout noise), error bars obtained from 100 independent Monte-Carlo trials with reported standard deviations, and the precise squeezing values and mode allocations used in each figure. A public code repository link is also provided. revision: yes

  2. Referee: [Proposed Method] The method computes PCA once on classical training data and then applies selective squeezing to those modes without re-deriving or validating the basis under the modified noise covariance induced by squeezing. Because squeezing changes the effective signal-to-noise structure, the classically dominant modes may no longer be optimal; the manuscript must either iterate the basis selection under the quantum noise model or demonstrate invariance of the subspace to support the resource-efficiency claim.

    Authors: This is a valid concern. We have added a new paragraph and accompanying figure in the revised manuscript that computes the principal components under the squeezed noise covariance and demonstrates substantial overlap (greater than 92 percent) with the classically obtained basis for the squeezing strengths considered. This invariance justifies the single-pass co-design for the reported operating regime. We also briefly discuss an iterative refinement procedure as a possible extension for stronger squeezing, although it is not required to achieve the claimed resource savings. revision: partial

Circularity Check

0 steps flagged

No circularity; co-design uses external classical PCA and numerical validation

full rationale

The paper proposes a methodological framework that computes PCA on classical training images to select a low-dimensional subspace, then applies squeezing only to those modes. This selection step is independent of the quantum noise covariance and does not reduce the reported accuracy or fidelity metrics to a fitted parameter or self-referential definition. No self-citations are invoked to justify uniqueness or to smuggle an ansatz. The central claims rest on comparative numerical experiments rather than an analytical derivation that loops back to its inputs by construction. The approach is therefore self-contained as a resource-allocation heuristic validated externally.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger is inferred from the high-level description. The approach relies on standard PCA and quantum optics but introduces the co-design choice without independent verification of its optimality.

free parameters (1)
  • number of principal components / squeezed modes
    The dimension of the subspace selected for squeezing is chosen to balance resource cost and performance; its value is not fixed by first principles.
axioms (1)
  • domain assumption Principal components computed from classical image statistics remain optimal for allocating quantum squeezing
    Invoked when the paper states that squeezing is applied selectively to the most informative spatial modes corresponding to these principal components.

pith-pipeline@v0.9.0 · 5488 in / 1301 out tokens · 26788 ms · 2026-05-10T08:09:10.481499+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

32 extracted references · 32 canonical work pages

  1. [1]

    torch.optim.lr_scheduler.OneCycleLR

    If the probe is prepared in a squeezed state with gainGaligned with the measured quadrature, the variance becomes Var( ˆXθ) = 1 2G < 1 2 ,(5) reducing measurement noise and improving the effec- tive SNR for estimating the image-dependent quadra- ture displacement. However, conventional pixel-wise squeezing requires dsqueezed states, so the resource cost s...

    work page

  2. [2]

    SVMs are supervised classifiers that learn a deci- sion boundary with maximum geometric margin

    Implementation details We use the SVMs to implement the classification. SVMs are supervised classifiers that learn a deci- sion boundary with maximum geometric margin. For binary-labeled training data{(x i, yi)}N i=1 withy i ∈ {±1}, a linear SVM seeks a separating hyperplane f(x) =w ⊤x+bby solving the soft-margin optimization minw,b,{ξi} 1 2 ∥w∥2 2 +C PN ...

    work page

  3. [3]

    (a) Accuracy of the MNIST data classification with k-PCA imaging versus the principal component numberk

    Experiment results (a) (b) Figure 5. (a) Accuracy of the MNIST data classification with k-PCA imaging versus the principal component numberk. Pixel numberm=dcovers the whole image. (b) Accuracy of the MNIST data classification withm-pixel15-PCA imag- ing versus the beamsplitter pixel numberm. Signal photon numbern S = 10per image. We show the accuracy cur...

    work page

  4. [4]

    Implementation details We use a convolutional neural network to implement end-to-end recovery from the measured principal com- ponent signals to the label of the object. We adopt a prescribed neural network design, residual network (ResNet) with 18 layers, equipped with dropout lay- ers to enhance generalizability, which is known to pro- duce >90% accurac...

    work page

  5. [5]

    Experiment results We show the accuracy curve of the CIFAR-10 dataset classification task in Fig. 6. Again, we observe signifi- cant quantum advantage with squeezingG= 10, which issignificantlybelowthenoiselesslimit, thusthereisstill a remarkable advantage margin that can be achieved by further increasing the squeezing gainG. Atk= 50, the quantum-enhanced...

    work page

  6. [6]

    Neural network architecture for denoising (decoder) withk-PCA input sizek= 50

    Implementation details Table I. Neural network architecture for denoising (decoder) withk-PCA input sizek= 50. Output is an image of size 28×28. Layer (type) Output shape Param # Linear-1[−1,1,128]6,528 Upsample-2[−1,128,3,3]0 ConvTranspose2d-3[−1,16,9,9]51,216 ConvTranspose2d-4[−1,8,21,21]3,208 ConvTranspose2d-5[−1,1,25,25]201 Upsample-6[−1,1,28,28]0 Tot...

    work page

  7. [7]

    Experiment results We present the image reconstruction results in Fig. 7. From top to bottom, we plot the truth, the classical benchmark with vacuum noise, and the quantum case with squeezed vacuum noise atG= 10. From left to right, we plot 10 examples of digit 0-9. We observe that the denoising completely fails in the classical case, while the quantum-en...

    work page

  8. [8]

    Journal of the Optical Society of America A3(8) (1986) 1152–1158

    Cox, c.I., Sheppard, C.: Information capacity and res- olution in an optical system. Journal of the Optical Society of America A3(8) (1986) 1152–1158

    work page 1986

  9. [9]

    Physical Review X6(3) (2016) 031033

    Tsang, M., Nair, R., Lu, X.M.: Quantum theory of superresolution for two incoherent optical point sources. Physical Review X6(3) (2016) 031033

    work page 2016

  10. [10]

    Physical Review Applied24(3) (2025) 034076

    Shi, H., Jones, C.M., Yu, M., Zhang, Z., Zhuang, Q.: Quantum-enhanced radio-frequency photonic dis- tributed imaging. Physical Review Applied24(3) (2025) 034076

    work page 2025

  11. [11]

    ACM Transactions on Graphics (TOG)37(4) (2018) 1– 13

    Sitzmann, V., Diamond, S., Peng, Y., Dun, X., Boyd, S., Heidrich, W., Heide, F., Wetzstein, G.: End-to-endopti- mization of optics and image processing for achromatic extended depth of field and super-resolution imaging. ACM Transactions on Graphics (TOG)37(4) (2018) 1– 13

    work page 2018

  12. [12]

    Advances in Neural Infor- mation Processing Systems29(2016)

    Chakrabarti, A.: Learning sensor multiplexing design through back-propagation. Advances in Neural Infor- mation Processing Systems29(2016)

    work page 2016

  13. [13]

    Physical Review D23(8) (1981) 1693–1708

    Caves, C.M.: Quantum-mechanical noise in an interfer- ometer. Physical Review D23(8) (1981) 1693–1708

    work page 1981

  14. [14]

    Journal of Modern Optics34(6-7) (1987) 881–902

    Collett, M., Loudon, R., Gardiner, C.: Quantum theory of optical homodyne and heterodyne detection. Journal of Modern Optics34(6-7) (1987) 881–902

    work page 1987

  15. [15]

    Reviews of Modern Physics84(2) (2012) 621–669

    Weedbrook, C., Pirandola, S., García-Patrón, R., Cerf, N.J., Ralph, T.C., Shapiro, J.H., Lloyd, S.: Gaussian quantum information. Reviews of Modern Physics84(2) (2012) 621–669

    work page 2012

  16. [16]

    Nature Photonics7(8) (2013) 613–619

    Aasi, J., Abadie, J., Abbott, B., Abbott, R., Abbott, T., Abernathy, M., Adams, C., Adams, T., Addesso, P., Adhikari, R., et al.: Enhanced sensitivity of the ligo gravitational wave detector by using squeezed states of light. Nature Photonics7(8) (2013) 613–619

    work page 2013

  17. [17]

    Physical Review Letters123(2019) 231107

    Tse, M.E., et al.: Quantum-enhanced advanced LIGO detectors in the era of gravitational-wave astronomy. Physical Review Letters123(2019) 231107

    work page 2019

  18. [18]

    PhysicalReviewLetters 123(2019) 231108

    Acernese, F., et al.: Increasing the astrophysical reach of the advanced Virgo detector via the application of squeezedvacuumstatesoflight. PhysicalReviewLetters 123(2019) 231108

    work page 2019

  19. [19]

    Nature Communications12(1) (2021) 4781

    Yang, Z., Jahanbozorgi, M., Jeong, D., Sun, S., Pfister, O., Lee, H., Yi, X.: A squeezed quantum microcomb on a chip. Nature Communications12(1) (2021) 4781

    work page 2021

  20. [20]

    Physical Review X15(4) (2025) 041009

    Hariri, A., Liu, S., Shi, H., Zhuang, Q., Fan, X., Zhang, Z.: Entangled dual-comb spectroscopy. Physical Review X15(4) (2025) 041009

    work page 2025

  21. [21]

    Physical Review Letters117(11) (2016) 110801

    Vahlbruch, H., Mehmet, M., Danzmann, K., Schnabel, R.: Detection of 15 db squeezed states of light and their application for the absolute calibration of photoelectric quantum efficiency. Physical Review Letters117(11) (2016) 110801

    work page 2016

  22. [22]

    IEEE Transactions on Information Theory52(4) (2006) 1289–1306

    Donoho, D.L.: Compressed sensing. IEEE Transactions on Information Theory52(4) (2006) 1289–1306

    work page 2006

  23. [23]

    IEEE Transactions on Information Theory52(2) (2006) 489–509

    Candes, E.J., Romberg, J., Tao, T.: Robust uncertainty principles: Exact signal reconstruction from highly in- complete frequency information. IEEE Transactions on Information Theory52(2) (2006) 489–509

    work page 2006

  24. [24]

    Candes, E.J., Tao, T.: Near-optimal signal recovery from random projections: Universal encoding strate- gies? IEEE Transactions on Information Theory52(12) (2006) 5406–5425

    work page 2006

  25. [25]

    ACS Photonics6(6) (2019) 1307–1318

    Marino, A.M., Pooser, R.C.: Quantum sensing with squeezed light. ACS Photonics6(6) (2019) 1307–1318

    work page 2019

  26. [26]

    Physical Review Applied14(6) (2020) 064026

    Banchi, L., Zhuang, Q., Pirandola, S.: Quantum- enhanced barcode decoding and pattern recognition. Physical Review Applied14(6) (2020) 064026

    work page 2020

  27. [27]

    Physical Review Applied20(2) (2023) 024072

    Ortolano, G., Napoli, C., Harney, C., Pirandola, S., Leonetti, G., Boucher, P., Losero, E., Genovese, M., Ruo-Berchera, I.: Quantum-enhanced pattern recog- nition. Physical Review Applied20(2) (2023) 024072

    work page 2023

  28. [28]

    Philosophical Magazine2(11) (1901) 559–572

    Pearson, K.: On lines and planes of closest fit to systems of points in space. Philosophical Magazine2(11) (1901) 559–572

    work page 1901

  29. [29]

    Journal of Educational Psychology24(6) (1933) 417–441

    Hotelling, H.: Analysis of a complex of statistical vari- ables into principal components. Journal of Educational Psychology24(6) (1933) 417–441

    work page 1933

  30. [30]

    Psychometrika1(3) (1936) 211–218

    Eckart, C., Young, G.: Theapproximationofonematrix by another of lower rank. Psychometrika1(3) (1936) 211–218

    work page 1936

  31. [31]

    springer, new york

    Jolliffe, I.: Principal component analysis, springer series in statistics. springer, new york. (2002)

    work page 2002

  32. [32]

    Physical Review Letters124(15) (2020) 150502

    Xia, Y., Li, W., Clark, W., Hart, D., Zhuang, Q., Zhang, 10 Z.: Demonstration of a reconfigurable entangled radio- frequency photonic sensor network. Physical Review Letters124(15) (2020) 150502

    work page 2020