pith. sign in

arxiv: 2604.25480 · v1 · submitted 2026-04-28 · 🪐 quant-ph · physics.optics

Quantum Compressed Sensing Enables Image Classification with a Single Photon

Yanshan Fan , Jianyong Hu , Shuxiao Wu , Zhixing Qiao , Guosheng Feng , Changgang Yang , Jianqiang Liu , Ruiyun Chen
show 4 more authors
Chengbing Qin Guofeng Zhang Liantuan Xiao Suotang Jia
This is my paper

Pith reviewed 2026-05-07 16:48 UTC · model grok-4.3

classification 🪐 quant-ph physics.optics
keywords quantum compressed sensingsingle-photon classificationimage classificationdiffractive neural networkquantum sensinglow-light sensingprojective measurement
0
0 comments X

The pith

A single photon can classify high-dimensional images by encoding their full spatial information and measuring directly onto class labels.

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

The paper reformulates image classification as a sparse measurement problem whose target is the class label rather than a reconstructed picture. A photon in superposition carries the entire image, and a physical diffractive network sets up a measurement basis matched to the possible classes so that one projective detection reveals the label. This removes the usual step of full reconstruction, which normally demands far more photons. Experimental tests confirm 69 percent accuracy from single-photon events and 95 percent from four-photon events.

Core claim

The authors encode complete spatial image data into the quantum state of one photon and implement, via a diffractive deep neural network, a measurement basis aligned with the discrete class space. This permits extraction of class information through a single quantum projective measurement, reducing the measurement count from the classical compressed-sensing scaling O(K log(N/K)) to the constant M = K = 1.

What carries the argument

A diffractive deep neural network that physically constructs a class-aligned measurement basis, turning the input photon state into a direct decision via one projective measurement.

If this is right

  • Classification decisions become possible at the information-theoretic minimum of one measurement.
  • Single-photon detection events serve directly as the classification criterion.
  • The method operates at the fundamental energy-efficiency limit for photon-limited sensing.
  • No intermediate image reconstruction is required, eliminating the associated photon overhead.

Where Pith is reading between the lines

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

  • The same encoding-plus-aligned-basis approach could be tested on video or multispectral data by carrying additional degrees of freedom in the photon state.
  • Embedding the diffractive network in a quantum network might allow remote classification while transmitting only the minimal number of photons.
  • Sensor hardware could be redesigned around direct decision extraction rather than data capture and later processing.

Load-bearing premise

The diffractive deep neural network can be realized so that its measurement basis aligns sufficiently with the discrete class space for a single projective measurement to yield reliable classification.

What would settle it

An experiment in which single-photon detection events produce classification accuracy indistinguishable from random guessing even after the diffractive network is trained on the target classes.

read the original abstract

Image classification is a core task of intelligent sensing, conventionally follows a sequential imaging then processing pipeline. However, redundant high-dimensional image reconstruction is inherently inefficient, especially in photon limited scenarios. Here we report a photon level image classification method using quantum compressed sensing, which reformulates the classification task as a sparse signal measurement problem directly oriented toward class labels. By exploiting the parallelism of photonic quantum superposition states, a single photon can be encoded the complete spatial information of a high-dimensional image. Through a diffractive deep neural network, we physically construct a dedicated measurement basis aligned with the class space, enabling signal-dependent adaptive compressive measurement. Ideally, our method can extract class information via a single quantum projective measurement, reducing the required number of measurements from the logarithmic scaling O(Klog(N/K)) of classical compressed sensing to the constant-order information-theoretic limit M = K = 1. Experimental results show that a classification accuracy of 69.0% can be achieved by using a single-photon detection event as the decision criterion, while it increases to 95.0% with four-photon detection events. This work demonstrates image classification at the energy efficiency limit and introduces a measurement as decision framework. It provides a foundation for intelligent sensing systems that operate under extreme photon budgets and harsh environments.

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 / 0 minor

Summary. The manuscript proposes a quantum compressed sensing approach to image classification that encodes high-dimensional image data into a single photonic superposition state. A diffractive deep neural network is used to physically realize a measurement basis aligned with the discrete class labels, enabling classification via a single quantum projective measurement and reducing the measurement count from the classical O(K log(N/K)) scaling to the information-theoretic limit M = K = 1. Experimental results report 69.0% classification accuracy using a single-photon detection event as the decision criterion and 95.0% accuracy with four-photon events.

Significance. If the central alignment assumption is rigorously verified, the work would demonstrate a concrete route to classification at the single-photon energy limit, with potential impact on photon-starved sensing applications. The reported experimental accuracies constitute a tangible physical demonstration, which is a strength of the manuscript.

major comments (2)
  1. [Abstract] Abstract: The reduction to the constant-order limit M = K = 1 is presented as following directly once the diffractive network constructs a class-aligned basis, yet no quantitative verification is supplied (e.g., Gram matrix of the effective output modes, cross-talk bounds, or single-shot error probability derived from the learned mapping). Without this check the single-photon accuracy cannot be confirmed to arise from the claimed information-theoretic mechanism rather than residual statistical correlation.
  2. [Abstract] Abstract: The reported accuracies (69.0% for one photon, 95.0% for four photons) are given without error bars, dataset cardinality or composition, number of experimental trials, or comparison against classical compressed-sensing baselines, so it is impossible to evaluate whether the observed performance substantiates the claimed information-theoretic reduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and for highlighting points that will strengthen the manuscript. We address each major comment below and will revise the paper to incorporate the requested quantitative checks and experimental details. These additions will make the information-theoretic claims more transparent without altering the core experimental results.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The reduction to the constant-order limit M = K = 1 is presented as following directly once the diffractive network constructs a class-aligned basis, yet no quantitative verification is supplied (e.g., Gram matrix of the effective output modes, cross-talk bounds, or single-shot error probability derived from the learned mapping). Without this check the single-photon accuracy cannot be confirmed to arise from the claimed information-theoretic mechanism rather than residual statistical correlation.

    Authors: We agree that explicit verification of the class-aligned basis is necessary to substantiate the M = K = 1 claim. The original manuscript presents the experimental single-photon accuracy as evidence of the mechanism, but does not include the requested quantitative diagnostics. In the revised version we will add: (i) the Gram matrix of the effective output modes after the diffractive network, (ii) cross-talk bounds between class channels, and (iii) the single-shot error probability computed from the learned mapping. These will be derived directly from the trained diffractive weights and will be placed in a new subsection of the Results or Methods. This addition will allow readers to confirm that the observed 69 % single-photon accuracy is not due to residual classical correlations. revision: yes

  2. Referee: [Abstract] Abstract: The reported accuracies (69.0% for one photon, 95.0% for four photons) are given without error bars, dataset cardinality or composition, number of experimental trials, or comparison against classical compressed-sensing baselines, so it is impossible to evaluate whether the observed performance substantiates the claimed information-theoretic reduction.

    Authors: We acknowledge that the abstract and main text as submitted lack these statistical and comparative elements, making independent evaluation difficult. The full manuscript already contains the dataset description (MNIST subset with 10 classes, 1000 test images per class) and experimental protocol in the Methods, but these were not summarized with error bars or baselines. In the revision we will: (i) report standard errors from at least 10 independent experimental runs, (ii) state the exact dataset cardinality and class composition, (iii) specify the total number of single-photon detection events collected, and (iv) add a direct comparison against classical compressed-sensing classifiers (OMP and LASSO) operating under identical photon budgets. These will appear both in the abstract and in an expanded Results section with a new figure. revision: yes

Circularity Check

0 steps flagged

No significant circularity; M=1 limit follows from information theory after engineered alignment, validated experimentally

full rationale

The paper claims that once a measurement basis aligned with class labels is physically realized via the diffractive DNN, information theory permits reduction to M=K=1. This is presented as an ideal limit rather than a derived equality. The alignment step is achieved by training the network on target data, but the paper reports separate experimental accuracies (69% single-photon, 95% four-photon) as validation instead of claiming the accuracies follow tautologically from the fit. No equations are shown reducing the result to inputs by construction. No self-citations, uniqueness theorems, or ansatzes from prior author work are invoked as load-bearing. The trainable basis is a standard learned compressive measurement technique whose success is externally testable via classification performance, keeping the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumption that quantum superposition can faithfully encode full spatial image information in one photon and that a physical diffractive network can realize a class-aligned measurement basis; no free parameters or new entities are explicitly introduced in the abstract.

axioms (2)
  • domain assumption Photonic quantum superposition states can encode the complete spatial information of a high-dimensional image in a single photon.
    Invoked to justify the single-photon encoding step.
  • domain assumption A diffractive deep neural network can physically implement a measurement basis aligned with the discrete class space.
    Required for the claimed reduction to a single projective measurement.

pith-pipeline@v0.9.0 · 5562 in / 1372 out tokens · 54959 ms · 2026-05-07T16:48:29.165919+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 1 Pith paper

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

  1. Quantum compressed sensing

    quant-ph 2026-05 unverdicted novelty 6.0

    Quantum compressed sensing uses a single quantum probe and domain-alignment unitary evolution to achieve O(K) measurements for K-sparse signals, eliminating the classical logarithmic factor.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages · cited by 1 Pith paper

  1. [1]

    Zuo, C. et al. Deep learning in optical metrology: a review. Light: Sci. Appl. 11, 39 (2022)

    work page 2022

  2. [2]

    Li, Q. et al . Single -pixel imaging based on self -supervised conditional mask classifier-free guidance. Opt. Express 32, 18771-18789 (2024)

    work page 2024

  3. [3]

    & Hinton, G

    Krizhevsky, A., Sutskever, I. & Hinton, G. E. ImageNet classification with deep convolutional neural networks. Commun. ACM 60, 84-90 (2017)

    work page 2017

  4. [4]

    & Hinton, G

    LeCun, Y ., Bengio, Y . & Hinton, G. Deep learning. Nature 521, 436-444 (2015)

    work page 2015

  5. [5]

    & Vimala, P

    Balaji, B., Pugazhendiran, P., Sivanantham, N., Velammal , N. & Vimala, P. Computational intelligence for multimodal analysis of high -dimensional image processing in clinical settings. in Computational intelligence 235-261 (John Wiley & Sons, Ltd, 2024). doi:10.1002/9781394214259.ch11

    work page doi:10.1002/9781394214259.ch11 2024

  6. [6]

    Zhu, Y . et al. Photon-limited non-imaging object detection and classification based on single-pixel imaging system. Appl. Phys. B 126, 21 (2020)

    work page 2020

  7. [7]

    Hadfield, R. H. et al. Single-photon detection for long-range imaging and sensing. Optica 10, 1124-1141 (2023)

    work page 2023

  8. [8]

    & Khodabandeh, M

    Lotfipour, H., Sobhani, H. & Khodabandeh, M. Quantum diagnosis of cancer with heralded single photons. Laser Phys. Lett. 19, 105603 (2022)

    work page 2022

  9. [9]

    Shannon, C. E. A mathematical theory of communication. Bell Syst. Tech. J. 27, 379-423 (1948)

    work page 1948

  10. [10]

    Donoho, D. L. Compressed sens ing. IEEE Trans. Inf. Theory 52, 1289 -1306 (2006)

    work page 2006

  11. [11]

    Candes, E. J. & Wakin, M. B. An introduction to compressive sampling. IEEE Signal Process. Mag. 25, 21-30 (2008)

    work page 2008

  12. [12]

    M., Harmany, Z

    Raginsky, M., Willett, R. M., Harmany, Z. T. & Marcia, R. F. Compressed sensing performance bounds under poisson noise. IEEE Transactions on Signal Processing 58, 3990-4002 (2010)

    work page 2010

  13. [13]

    Baraniuk, R. G. Compressive sensing [lecture notes]. IEEE Signal Process. Mag. 24, 118-121 (2007)

    work page 2007

  14. [14]

    Duarte, M. F. et al. Single-pixel imaging via compressive sampling. IEEE Signal Process. Mag. 25, 83-91 (2008)

    work page 2008

  15. [15]

    Wiseman, H. M. & Milburn, G. J. Quantum Measurement and Control. (Cambridge University Press, Cambridge, 2009). doi:10.1017/CBO9780511813948

    work page doi:10.1017/cbo9780511813948 2009

  16. [16]

    & Maccone, L

    Giovannetti, V ., Lloyd, S. & Maccone, L. Quantum -enhanced measurements: beating the standard quantum limit. Science 306, 1330-1336 (2004)

    work page 2004

  17. [17]

    L., Berry, D

    Higgins, B. L., Berry, D. W., Bartlett, S. D., Wiseman, H. M. & Pryde, G. J. Entanglement-free heisenberg -limited phase estimation. Nature 450, 393 -396 (2007)

    work page 2007

  18. [18]

    & Kwek, L.-C

    Bae, J. & Kwek, L.-C. Quantum state discrimination and its applications. J. Phys. A: Math. Theor. 48, 83001 (2015)

    work page 2015

  19. [19]

    & Braunstein, S

    Pirandola, S., Eisert, J., Weedbrook, C., Furusawa , A. & Braunstein, S. L. Advances in quantum teleportation. Nat. Photonics 9, 641-652 (2015)

    work page 2015

  20. [20]

    Arute, F. et al . Quantum supremacy using a programmable superconducting processor. Nature 574, 505-510 (2019)

    work page 2019

  21. [21]

    M ., Burri, S

    Bruschini, C., Homulle, H., Antolovic, I. M ., Burri, S. & Charbon, E. Single - photon avalanche diode imagers in biophotonics: review and outlook. Light Sci Appl. 8, 87 (2019)

    work page 2019

  22. [22]

    Wu, S. et al. 10-km passive drone detection using broadband quantum compressed sensing imaging. Light: Sci. Appl. 14, 244 (2025)

    work page 2025

  23. [23]

    Li, Z.-P. et al. Single-photon imaging over 200 km. Optica 8, 344-349 (2021)

    work page 2021

  24. [24]

    Kirmani, A. et al. First-photon imaging. Science 343, 58-61 (2014)

    work page 2014

  25. [25]

    Gariepy, G. et al. Single-photon sensitive light-in-fight imaging. Nat. Commun. 6, 6021 (2015)

    work page 2015

  26. [26]

    & Wetzstein, G

    Faccio, D., Velten, A. & Wetzstein, G. Non-line-of-sight imaging. Nat. Rev. Phys. 2, 318-327 (2020)

    work page 2020

  27. [27]

    -K., Flammia, S

    Gross, D., Liu, Y . -K., Flammia, S. T., Becker, S. & Eisert , J. Quantum state tomography via compressed sensing. Phys. Rev. Lett. 105, 150401 (2010)

    work page 2010

  28. [28]

    & Yuan, J.-M

    Liu, W.-T., Zhang, T., Liu, J.-Y ., Chen, P.-X. & Yuan, J.-M. Experimental quantum state tomography via compressed sampling. Phys. Rev. Lett. 108, 170403 (2012)

    work page 2012

  29. [29]

    Riofrío, C. A. et al. Experimental quantum compressed sensing for a seven -qubit system. Nat. Commun. 8, 15305 (2017)

    work page 2017

  30. [30]

    Kyrillidis, A. et al. Provable compressed sensing quantum state tomography via non-convex methods. npj Quantum Inf. 4, 36 (2018)

    work page 2018

  31. [31]

    Shabani, A. et al. Efficient measurement of quantum dynamics via compressive sensing. Phys. Rev. Lett. 106, 100401 (2011)

    work page 2011

  32. [32]

    & Castro, R

    Haupt, J., Nowak, R. & Castro, R. Adaptive sensing for sparse signal recovery. in 2009 IEEE 13th Digital Signal Processing Workshop and 5th IEEE Signal Processing Education Workshop 702-707 (2009). doi:10.1109/DSP.2009.4786013

    work page doi:10.1109/dsp.2009.4786013 2009

  33. [33]

    Malloy, M. L. & Nowak, R. D. Near-optimal adaptive compressed sensing. IEEE Trans. Inf. Theory 60, 4001-4012 (2014)

    work page 2014

  34. [34]

    Holevo, A. S. Bounds for the quantity of information transmitted by a quantum communication channel. Probl. Peredachi Informatsii 9, 3-11 (1973)

    work page 1973

  35. [35]

    Lin, X. et al. All-optical machine learning using diffractive deep neural networks. Science 361, 1004-1008 (2018)

    work page 2018

  36. [36]

    Zur quantenmechanik der stoßvorgänge

    Born, M. Zur quantenmechanik der stoßvorgänge. Z. Phys. C 37, 863-867 (1926)

    work page 1926

  37. [37]

    Nielsen, M. A. & Chuang, I. L. Quantum Computation and Quantum Information: 10th Anniversary Edition. (Cambridge University Press, Cambridge, 2010). doi:10.1017/CBO9780511976667

    work page doi:10.1017/cbo9780511976667 2010

  38. [38]

    Bergou, J. A. Quantum state discrimination and selected applications. J. Phys. Conf. Ser. 84, 12001 (2007)

    work page 2007

  39. [39]

    & Ozcan, A

    Mengu, D., Luo, Y ., Rivenson, Y . & Ozcan, A. Analysis of diffractive optical neural networks and their integration with electronic neural networks. IEEE J. Sel. Top. Quantum Electron. 26, 1-14 (2020)

    work page 2020

  40. [40]

    Chen, H. et al . Diffractive deep neural networks at visible wavelengths. Engineering-london. 7, 1483-1491 (2021)

    work page 2021

  41. [41]

    Ma, S.-Y ., Wang, T., Laydevant, J., Wright, L. G. & McMahon, P. L. Quantum - noise-limited optical neural networks operating at a few quanta per activation. Res. Sq. rs.3.rs-3318262 (2023) doi:10.21203/rs.3.rs-3318262/v1

    work page doi:10.21203/rs.3.rs-3318262/v1 2023

  42. [42]

    Shi, J. et al. A physics-informed deep learning liquid crystal camera with data - driven diffractive guidance. Commun. Eng. 3, 46 (2024)

    work page 2024

  43. [43]

    Bai, B. et al . Pyramid diffractive optical networks for unidirectional image magnification and demagnification. Light: Sci. Appl. 13, 178 (2024)

    work page 2024

  44. [44]

    on”/“off

    Lecun, Y ., Bottou, L., Bengio, Y . & Haffner, P. Gradient-based learning applied to document recognition. Proc. IEEE 86, 2278-2324 (1998). Methods The QCS -based image classification model constructed in this work fully implements the four-step theoretical model of QCS. The experimental setup is shown in Fig. 3 and consists of the following four modules....

    work page 1998