pith. sign in

arxiv: 2604.03482 · v1 · submitted 2026-04-03 · 🪐 quant-ph

Learning high-dimensional quantum entanglement through physics-guided neural networks

Yang Xu , Hao Zhang , Wenwen Zhang , Luchang Niu , Girish Kulkarni , Mahtab Amooei , Sergio Carbajo , Robert W. Boyd This is my paper

Pith reviewed 2026-05-13 18:23 UTC · model grok-4.3

classification 🪐 quant-ph
keywords high-dimensionalentanglementmodalphysics-guidedtrainingacrosscharacterizationfull
0
0 comments X

The pith

Physics-guided neural networks reconstruct the high-dimensional modal structure of SPDC entanglement with high fidelity and 128-fold speedup over simulation.

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

The paper proposes a neural network that reconstructs the joint distribution of entangled photon modes across radial and azimuthal indices in high-gain spontaneous parametric down-conversion. A sympathetic reader would care because direct numerical simulation of the full multimode output is too slow for practical experimental use, blocking real-time characterization of these bright entangled states. The method embeds physical knowledge by training on simulated targets with a hybrid loss that includes a soft orbital-angular-momentum conservation penalty, producing predictions that remain consistent even with limited or noisy data. Results show average JSD of 1.96e-3 and more than 30 percent accuracy improvement over standard U-Net models while delivering an approximate 128-fold computational speedup.

Core claim

We designed a FiLM-modulated convolutional architecture that predicts the joint (m,l) distribution, and training is driven by a hybrid loss that couples data-driven metrics (JSD, KL, MSE, Wasserstein) with a soft orbital-angular-momentum conservation term, providing an essential inductive bias toward physically consistent solutions. Across gain regimes, our method achieves high-fidelity reconstruction with average JSD of 1.96e-3, WEMD of 1.54e-3, and KL divergence of 7.85e-3, delivering an approximate 128-fold speedup over full numerical simulation and more than 30% accuracy gains over U-Net baselines. These results demonstrate that physics-guided learning, via a soft OAM-conservationregular

What carries the argument

FiLM-modulated convolutional network trained with hybrid loss containing a soft orbital-angular-momentum conservation regularizer to predict joint (m,l) modal distributions

Where Pith is reading between the lines

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

  • The method could enable real-time experimental tuning of entanglement sources by providing immediate modal feedback during data collection.
  • It may generalize to other multimode nonlinear optical processes that produce high-dimensional quantum states.
  • Reducing the computational cost could allow systematic exploration of entanglement properties across wider parameter ranges than currently feasible.
  • The soft regularizer approach might be combined with other physical symmetries to improve data efficiency in related quantum optics tasks.

Load-bearing premise

Training solely on simulated SPDC data plus a soft OAM conservation term yields predictions that stay accurate when the network encounters real experimental data whose noise and imperfections differ from the training distribution.

What would settle it

Apply the trained network to measured modal data from an actual high-gain SPDC experiment and compare its predicted joint (m,l) distribution against independent full numerical simulation or direct tomography of the same experimental run.

Figures

Figures reproduced from arXiv: 2604.03482 by Girish Kulkarni, Hao Zhang, Luchang Niu, Mahtab Amooei, Robert W. Boyd, Sergio Carbajo, Wenwen Zhang, Yang Xu.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
read the original abstract

High-gain spontaneous parametric down-conversion (SPDC) produces bright squeezed vacuum with rich high-dimensional entanglement, but its output is inherently multimodal and non-perturbative, making the full modal characterization a major computational bottleneck. We propose a physics-guided deep neural network that reconstructs the source's modal fingerprint: the high-dimensional correlation signature across radial and azimuthal indices. We designed a FiLM-modulated convolutional architecture that predicts the joint (m,l) distribution, and training is driven by a hybrid loss that couples data-driven metrics (JSD, KL, MSE, Wasserstein) with a soft orbital-angular-momentum (OAM) conservation term, providing an essential inductive bias toward physically consistent solutions. Across gain regimes, our method achieves high-fidelity reconstruction with average JSD of 1.96e-3, WEMD of 1.54e-3, and KL divergence of 7.85e-3, delivering an approximate 128-fold speedup over full numerical simulation and more than 30% accuracy gains over U-Net baselines. These results demonstrate that physics-guided learning, via a soft OAM-conservation regularizer and physically generated training targets, enables rapid and data-efficient modal characterization. Compared with traditional numerical simulation, our mesh-free method has demonstrated good generalization with limited or contaminated training data and has enabled fast "online" prediction of the quantum dynamics of a high-dimensional entanglement system for real-world experimental implementation.

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 introduces a FiLM-modulated convolutional neural network trained on simulated high-gain SPDC data to reconstruct the joint (m,l) modal distribution of high-dimensional entanglement. A hybrid loss combines standard distributional metrics (JSD, KL, MSE, Wasserstein) with a soft OAM-conservation regularizer. Across gain regimes the network reports average JSD of 1.96e-3, WEMD of 1.54e-3 and KL of 7.85e-3, together with a claimed 128-fold speedup over full numerical simulation and >30 % accuracy improvement over U-Net baselines. The authors conclude that the physics-guided approach enables rapid, data-efficient modal characterization suitable for real-world experimental implementation.

Significance. If the reported metrics and sim-to-real transfer hold, the method would remove a major computational bottleneck in characterizing bright squeezed vacuum, permitting online prediction of entanglement structure during experiments. The explicit incorporation of an OAM soft constraint as an inductive bias is a clear strength that distinguishes the work from purely data-driven baselines.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (Results): the headline performance figures and the claim of suitability for 'real-world experimental implementation' rest entirely on simulated SPDC targets; no experimental data, noise-model mismatch tests, or sim-to-real transfer experiments are presented. This directly undermines the generalization statement that is load-bearing for the central contribution.
  2. [§3.2] §3.2 (Loss function): the hybrid-loss coefficients are treated as free parameters yet no ablation, sensitivity analysis, or selection protocol is reported. Because these weights control the balance between data fidelity and the OAM constraint, their arbitrary choice affects whether the quoted JSD/WEMD/KL values are reproducible or merely tuned.
minor comments (2)
  1. [Figure 3 and §4.1] Figure 3 caption and §4.1: the speedup factor of 128× is stated without specifying the hardware baseline, mesh resolution, or whether the comparison includes data-generation time; a precise timing table would clarify the practical gain.
  2. Notation: the symbols for radial and azimuthal indices are introduced inconsistently (sometimes (m,l), sometimes (l,m)); a single consistent convention should be adopted throughout.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We have revised the manuscript to address the concerns about experimental validation and loss-function hyperparameters. Point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (Results): the headline performance figures and the claim of suitability for 'real-world experimental implementation' rest entirely on simulated SPDC targets; no experimental data, noise-model mismatch tests, or sim-to-real transfer experiments are presented. This directly undermines the generalization statement that is load-bearing for the central contribution.

    Authors: We agree that the reported results rely on simulated targets. In the revised version we have updated the abstract and §4 to state explicitly that all quantitative metrics are obtained from physics-based simulations and to moderate the language on immediate experimental deployment. We have added a new subsection in §4 that includes robustness tests under additive Gaussian noise and simulated beam misalignment (mimicking typical experimental imperfections), with the network retaining JSD < 5e-3. A brief discussion of domain-adaptation strategies for future sim-to-real transfer has also been included. These changes clarify the current scope while preserving the core claim that the physics-guided architecture offers a practical route to fast modal reconstruction. revision: partial

  2. Referee: [§3.2] §3.2 (Loss function): the hybrid-loss coefficients are treated as free parameters yet no ablation, sensitivity analysis, or selection protocol is reported. Because these weights control the balance between data fidelity and the OAM constraint, their arbitrary choice affects whether the quoted JSD/WEMD/KL values are reproducible or merely tuned.

    Authors: We thank the referee for highlighting this omission. The revised §3.2 now contains an ablation study in which each loss coefficient is varied independently over [0, 1] while the others are held fixed; the resulting JSD, KL, WEMD, and OAM-violation metrics are reported. The original weights were obtained by a grid search on a validation split that minimized JSD subject to an OAM-conservation error threshold. The ablation demonstrates that performance remains stable within approximately ±20 % of the chosen values, indicating that the quoted metrics are reproducible rather than the result of a single arbitrary tuning. revision: yes

Circularity Check

0 steps flagged

No circularity: reconstruction targets generated by independent SPDC simulations; OAM regularizer is external domain constraint

full rationale

The paper trains a FiLM-modulated CNN to predict the joint (m,l) modal distribution using targets produced by separate numerical simulations of high-gain SPDC. The hybrid loss combines standard distributional metrics (JSD, KL, MSE, Wasserstein) with a soft OAM-conservation penalty derived from angular-momentum selection rules; this penalty acts as an inductive bias rather than redefining the target quantity. Reported averages (JSD 1.96e-3, WEMD 1.54e-3, KL 7.85e-3) are therefore direct comparisons against held-out simulated ground truth, not quantities that reduce to the network parameters by construction. The 128-fold speedup is a wall-clock comparison to full numerical integration and does not rely on self-citation chains or ansatz smuggling. No load-bearing step equates the claimed output to its own inputs; the derivation remains a standard supervised physics-informed learning pipeline.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that numerical SPDC simulations produce faithful training targets and that a soft OAM conservation penalty is sufficient to enforce physical consistency without introducing new free parameters that define the result.

free parameters (1)
  • hybrid loss coefficients
    Relative weights among JSD, KL, MSE, Wasserstein, and OAM terms must be chosen during training; their specific values are not stated.
axioms (1)
  • domain assumption Orbital angular momentum is conserved in the SPDC interaction
    Invoked as a soft constraint inside the hybrid loss to bias network outputs toward physically allowed solutions.

pith-pipeline@v0.9.0 · 5579 in / 1431 out tokens · 59082 ms · 2026-05-13T18:23:07.746987+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

70 extracted references · 70 canonical work pages

  1. [1]

    As a feasible solution, we employ a deep learning neu- ral network based on a Feature-wise Linear Modulation (FiLM)-conditioned residual architecture, as shown in Figure 2a

    The computation of the full modal structure and the Schmidt number of high-dimensional entanglement state involves multi-dimensional numerical integration and sin- gular value decomposition of large matrices, thus leading to extremely high time complexity (see Methods). As a feasible solution, we employ a deep learning neu- ral network based on a Feature-...

    work page

  2. [2]

    squeezing eigenmodes

    Numerical baseline simulation To numerically compute the full mode structure of the two-photon wavefunction (Eq. 2), and the Schmidt num- ber numerically, we use cylindrical coordinates for both signal and idler transverse wavevectors,q s andq i. In the cylindrical coordinate system, the rotational symmetry of 10 the system allows us to reduce the azimuth...

    work page

  3. [3]

    The driving pulse has a pulse width of 30 ps (FWHM) and a repetition rate of 50 Hz

    Experiment Setup In our experiment, we use a 355-nm vertically- polarized pulsed Nd:YAG laser (EKSPLA PL2231) to drive the SPDC process. The driving pulse has a pulse width of 30 ps (FWHM) and a repetition rate of 50 Hz. The driving pulse is first spatially-filtered and then sent to a 3-mm type-I BBO (β-barium borate) crystal (cut for type-I degenerate co...

    work page

  4. [4]

    P. G. Kwiat, A. M. Steinberg, and R. Y. Chiao, High- visibility interference in a bell-inequality experiment for energy and time, Phys. Rev. A47, R2472 (1993)

    work page 1993

  5. [5]

    D. V. Strekalov, T. B. Pittman, A. V. Sergienko, Y. H. Shih, and P. G. Kwiat, Postselection-free energy-time en- tanglement, Phys. Rev. A54, R1 (1996)

    work page 1996

  6. [6]

    C. K. Law and J. H. Eberly, Analysis and interpreta- tion of high transverse entanglement in optical paramet- 11 ric down conversion, Phys. Rev. Lett.92, 127903 (2004)

    work page 2004

  7. [7]

    S. P. Walborn and C. H. Monken, Transverse spatial en- tanglement in parametric down-conversion, Phys. Rev. A 76, 062305 (2007)

    work page 2007

  8. [8]

    M. H. Rubin, D. N. Klyshko, Y. H. Shih, and A. V. Sergienko, Theory of two-photon entanglement in type- ii optical parametric down-conversion, Phys. Rev. A50, 5122 (1994)

    work page 1994

  9. [9]

    Fiorentino, G

    M. Fiorentino, G. Messin, C. E. Kuklewicz, F. N. C. Wong, and J. H. Shapiro, Generation of ultrabright tun- able polarization entanglement without spatial, spectral, or temporal constraints, Phys. Rev. A69, 041801 (2004)

    work page 2004

  10. [10]

    A. Mair, A. Vaziri, G. Weihs, and A. Zeilinger, Entangle- ment of the orbital angular momentum states of photons, Nature412, 313 (2001)

    work page 2001

  11. [11]

    Fickler, R

    R. Fickler, R. Lapkiewicz, W. N. Plick, M. Krenn, C. Schaeff, S. Ramelow, and A. Zeilinger, Quantum en- tanglement of high angular momenta, Science338, 640 (2012)

    work page 2012

  12. [12]

    Romero, D

    J. Romero, D. Giovannini, S. Franke-Arnold, S. Bar- nett, and M. Padgett, Increasing the dimension in high- dimensional two-photon orbital angular momentum en- tanglement, Physical Review A86, 012334 (2012)

    work page 2012

  13. [13]

    F. M. Miatto, A. M. Yao, and S. M. Barnett, Full charac- terization of the quantum spiral bandwidth of entangled biphotons, Physical Review A83, 033816 (2011)

    work page 2011

  14. [14]

    B. J. Lawrie, P. D. Lett, A. M. Marino, and R. C. Pooser, Quantum sensing with squeezed light, Acs Photonics6, 1307 (2019)

    work page 2019

  15. [15]

    Brida, I

    G. Brida, I. P. Degiovanni, M. Genovese, M. L. Rastello, and I. Ruo-Berchera, Detection of multimode spatial cor- relation in pdc and application to the absolute calibration of a ccd camera, Optics Express18, 20572 (2010)

    work page 2010

  16. [16]

    Brida, M

    G. Brida, M. Genovese, and I. Ruo Berchera, Experi- mental realization of sub-shot-noise quantum imaging, Nature Photonics4, 227 (2010)

    work page 2010

  17. [17]

    Lopaeva, I

    E. Lopaeva, I. Ruo Berchera, I. P. Degiovanni, S. Oli- vares, G. Brida, and M. Genovese, Experimental realiza- tion of quantum illumination, Physical Review Letters 110, 153603 (2013)

    work page 2013

  18. [18]

    Hurvitz, A

    I. Hurvitz, A. Karnieli, and A. Arie, Frequency-domain engineering of bright squeezed vacuum for continuous- variable quantum information, Optics Express31, 20387 (2023)

    work page 2023

  19. [19]

    Sharapova, A

    P. Sharapova, A. M. P´ erez, O. V. Tikhonova, and M. V. Chekhova, Schmidt modes in the angular spectrum of bright squeezed vacuum, Physical Review A91, 043816 (2015)

    work page 2015

  20. [20]

    Iskhakov, M

    T. Iskhakov, M. V. Chekhova, and G. Leuchs, Gen- eration and direct detection of broadband mesoscopic polarization-squeezed vacuum, Phys. Rev. Lett.102, 183602 (2009)

    work page 2009

  21. [21]

    I. N. Agafonov, M. V. Chekhova, and G. Leuchs, Two- color bright squeezed vacuum, Phys. Rev. A82, 011801 (2010)

    work page 2010

  22. [22]

    Rasputnyi, Z

    A. Rasputnyi, Z. Chen, M. Birk, O. Cohen, I. Kaminer, M. Kr¨ uger, D. Seletskiy, M. Chekhova, and F. Tani, High- harmonic generation by a bright squeezed vacuum, Na- ture Physics20, 1960 (2024)

    work page 1960

  23. [23]

    Lemieux, S

    S. Lemieux, S. A. Jalil, D. N. Purschke, N. Boroumand, T. Hammond, D. Villeneuve, A. Naumov, T. Brabec, and G. Vampa, Photon bunching in high-harmonic emission controlled by quantum light, Nature Photonics , 1 (2025)

    work page 2025

  24. [24]

    Z. Lyu, F. Sun, Y. Fang, Q. He, and Y. Liu, Effect of pho- ton quantum statistics on electrons in above-threshold ionization, Physical Review Research7, L012072 (2025)

    work page 2025

  25. [25]

    Heimerl, A

    J. Heimerl, A. Rasputnyi, J. P¨ olloth, S. Meier, M. Chekhova, and P. Hommelhoff, Quantum light drives electrons strongly at metal needle tips, Nature Physics , 1 (2025)

    work page 2025

  26. [26]

    Sharapova, G

    P. Sharapova, G. Frascella, M. Riabinin, A. P´ erez, O. Tikhonova, S. Lemieux, R. Boyd, G. Leuchs, and M. Chekhova, Properties of bright squeezed vacuum at increasing brightness, Physical Review Research2, 013371 (2020)

    work page 2020

  27. [27]

    Xiao, L.-A

    M. Xiao, L.-A. Wu, and H. J. Kimble, Precision measure- ment beyond the shot-noise limit, Physical review letters 59, 278 (1987)

    work page 1987

  28. [28]

    R. C. Pooser and B. Lawrie, Ultrasensitive measure- ment of microcantilever displacement below the shot- noise limit, Optica2, 393 (2015)

    work page 2015

  29. [29]

    Oelker, G

    E. Oelker, G. Mansell, M. Tse, J. Miller, F. Matichard, L. Barsotti, P. Fritschel, D. McClelland, M. Evans, and N. Mavalvala, Ultra-low phase noise squeezed vacuum source for gravitational wave detectors, Optica3, 682 (2016)

    work page 2016

  30. [30]

    Cutipa and M

    P. Cutipa and M. V. Chekhova, Bright squeezed vacuum for two-photon spectroscopy: simultaneously high resolu- tion in time and frequency, space and wavevector, Optics Letters47, 465 (2022)

    work page 2022

  31. [31]

    Even Tzur and O

    M. Even Tzur and O. Cohen, Motion of charged parti- cles in bright squeezed vacuum, Light: Science & Appli- cations13, 41 (2024)

    work page 2024

  32. [32]

    F. M. Miatto, A. M. Yao, and S. M. Barnett, Full charac- terization of the quantum spiral bandwidth of entangled biphotons, Phys. Rev. A83, 033816 (2011)

    work page 2011

  33. [33]

    Keller, A

    A. Keller, A. Z. Khoury, N. Fabre, M. Amanti, F. Baboux, S. Ducci, and P. Milman, Reconstructing the full modal structure of photonic states by stimulated- emission tomography in the low-gain regime, Phys. Rev. A106, 063709 (2022)

    work page 2022

  34. [34]

    Barbastathis, A

    G. Barbastathis, A. Ozcan, and G. Situ, On the use of deep learning for computational imaging, Optica6, 921 (2019)

    work page 2019

  35. [35]

    Zhang, Y

    H. Zhang, Y. Xu, W. Zhang, S. Choudhary, M. Z. Alam, L. D. Nguyen, M. Klein, S. Vangala, J. K. Miller, E. G. Johnson,et al., Hybrid deep reconstruction for vignetting-free upconversion imaging through scatter- ing in enz materials, arXiv preprint arXiv:2508.13096 (2025)

    work page arXiv 2025

  36. [36]

    X. Lin, Y. Rivenson, N. T. Yardimci, M. Veli, Y. Luo, M. Jarrahi, and A. Ozcan, All-optical machine learning using diffractive deep neural networks, Science361, 1004 (2018)

    work page 2018

  37. [37]

    Zhang, T

    H. Zhang, T. Zhao, W. Zhang, S. Halvaei, M. Peters, T. S. Cheung, K. C. Williams, and R. Gordon, Accurate label free classification of cancerous extracellular vesicles using nanoaperture optical tweezers and deep learning, npj Biosensing2, 33 (2025)

    work page 2025

  38. [38]

    S. Yuan, C. Ma, E. Fetaya, T. Mueller, D. Naveh, F. Zhang, and F. Xia, Geometric deep optical sensing, Science379, eade1220 (2023)

    work page 2023

  39. [39]

    C. Zuo, J. Qian, S. Feng, W. Yin, Y. Li, P. Fan, J. Han, K. Qian, and Q. Chen, Deep learning in optical metrol- ogy: a review, Light: Science & Applications11, 39 (2022)

    work page 2022

  40. [40]

    Z. A. Kudyshev, V. M. Shalaev, and A. Boltasseva, Ma- chine learning for integrated quantum photonics, ACS 12 Photonics8, 34 (2020)

    work page 2020

  41. [41]

    W. Ji, J. Chang, H.-X. Xu, J. R. Gao, S. Gr¨ oblacher, H. P. Urbach, and A. J. Adam, Recent advances in meta- surface design and quantum optics applications with ma- chine learning, physics-informed neural networks, and topology optimization methods, Light: Science & Ap- plications12, 169 (2023)

    work page 2023

  42. [42]

    Genty, L

    G. Genty, L. Salmela, J. M. Dudley, D. Brunner, A. Kokhanovskiy, S. Kobtsev, and S. K. Turitsyn, Ma- chine learning and applications in ultrafast photonics, Nature Photonics15, 91 (2021)

    work page 2021

  43. [43]

    Salmela, N

    L. Salmela, N. Tsipinakis, A. Foi, C. Billet, J. M. Dudley, and G. Genty, Predicting ultrafast nonlinear dynamics in fibre optics with a recurrent neural network, Nature machine intelligence3, 344 (2021)

    work page 2021

  44. [44]

    Schmale, M

    T. Schmale, M. Reh, and M. G¨ arttner, Efficient quantum state tomography with convolutional neural networks, npj Quantum Information8, 115 (2022)

    work page 2022

  45. [45]

    Y. Quek, S. Fort, and H. K. Ng, Adaptive quantum state tomography with neural networks, npj Quantum Infor- mation7, 105 (2021)

    work page 2021

  46. [46]

    Ure˜ na, A

    J. Ure˜ na, A. Sojo, J. Bermejo-Vega, and D. Manzano, Entanglement detection with classical deep neural net- works, Scientific Reports14, 18109 (2024)

    work page 2024

  47. [47]

    D.-L. Deng, X. Li, and S. Das Sarma, Quantum entan- glement in neural network states, Physical Review X7, 021021 (2017)

    work page 2017

  48. [48]

    Youssry, Y

    A. Youssry, Y. Yang, R. J. Chapman, B. Haylock, F. Lenzini, M. Lobino, and A. Peruzzo, Experimental graybox quantum system identification and control, npj Quantum Information10, 9 (2024)

    work page 2024

  49. [49]

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

    work page 2019

  50. [50]

    A. Saba, C. Gigli, A. B. Ayoub, and D. Psaltis, Physics- informed neural networks for diffraction tomography, Ad- vanced Photonics4, 066001 (2022)

    work page 2022

  51. [51]

    Zhang, L

    H. Zhang, L. Sun, J. Hirschman, and S. Carbajo, Multi- modality deep learning for pulse prediction in homoge- neous nonlinear systems via parametric conversion, APL Photonics10(2025)

    work page 2025

  52. [52]

    Y. Tang, J. Fan, X. Li, J. Ma, M. Qi, C. Yu, and W. Gao, Physics-informed recurrent neural network for time dy- namics in optical resonances, Nature computational sci- ence2, 169 (2022)

    work page 2022

  53. [53]

    Momeni, B

    A. Momeni, B. Rahmani, B. Scellier, L. G. Wright, P. L. McMahon, C. C. Wanjura, Y. Li, A. Skalli, N. G. Berloff, T. Onodera,et al., Training of physical neural networks, Nature645, 53 (2025)

    work page 2025

  54. [54]

    Y. Chen, L. Lu, G. E. Karniadakis, and L. Dal Negro, Physics-informed neural networks for inverse problems in nano-optics and metamaterials, Optics express28, 11618 (2020)

    work page 2020

  55. [55]

    LeCun, K

    Y. LeCun, K. Kavukcuoglu, and C. Farabet, Convolu- tional networks and applications in vision, inProceedings of 2010 IEEE International Symposium on Circuits and Systems(IEEE, 2010) pp. 253–256

    work page 2010

  56. [56]

    Isensee, P

    F. Isensee, P. F. Jaeger, S. A. Kohl, J. Petersen, and K. H. Maier-Hein, nnu-net: a self-configuring method for deep learning-based biomedical image segmentation, Na- ture methods18, 203 (2021)

    work page 2021

  57. [57]

    Y. Chen, T. Zhou, J. Wu, H. Qiao, X. Lin, L. Fang, and Q. Dai, Photonic unsupervised learning variational autoencoder for high-throughput and low-latency image transmission, Science Advances9, eadf8437 (2023)

    work page 2023

  58. [58]

    Boussafa, L

    Y. Boussafa, L. Sader, V. T. Hoang, B. P. Chaves, A. Bougaud, M. Fabert, A. Tonello, J. M. Dudley, M. Kues, and B. Wetzel, Deep learning prediction of noise-driven nonlinear instabilities in fibre optics, Nature Communications16, 7800 (2025)

    work page 2025

  59. [59]

    F. M. Miatto, H. di Lorenzo Pires, S. M. Barnett, and M. P. van Exter, Spatial schmidt modes generated in parametric down-conversion, The European Physical Journal D66, 263 (2012)

    work page 2012

  60. [60]

    L. Kopf, R. Barros, S. Prabhakar, E. Giese, and R. Fickler, Conservation of angular momentum on a single-photon level, Physical Review Letters134, 203601 (2025)

    work page 2025

  61. [61]

    Vallone, V

    G. Vallone, V. D’Ambrosio, A. Sponselli, S. Slussarenko, L. Marrucci, F. Sciarrino, and P. Villoresi, Free-space quantum key distribution by rotation-invariant twisted photons, Phys. Rev. Lett.113, 060503 (2014)

    work page 2014

  62. [62]

    Feng and P

    S. Feng and P. Kumar, Spatial symmetry and conserva- tion of orbital angular momentum in spontaneous para- metric down-conversion, Phys. Rev. Lett.101, 163602 (2008)

    work page 2008

  63. [63]

    L. Kopf, R. Barros, S. Prabhakar, E. Giese, and R. Fick- ler, Conservation of angular momentum on a single- photon level, Phys. Rev. Lett.134, 203601 (2025)

    work page 2025

  64. [64]

    Kulkarni, R

    G. Kulkarni, R. Sahu, O. S. Maga˜ na-Loaiza, R. W. Boyd, and A. K. Jha, Single-shot measurement of the orbital- angular-momentum spectrum of light, Nature communi- cations8, 1054 (2017)

    work page 2017

  65. [65]

    Kulkarni, J

    G. Kulkarni, J. Rioux, B. Braverman, M. V. Chekhova, and R. W. Boyd, Classical model of spontaneous para- metric down-conversion, Physical Review Research4, 033098 (2022)

    work page 2022

  66. [66]

    Z. Hao, Z. Wang, H. Su, C. Ying, Y. Dong, S. Liu, Z. Cheng, J. Song, and J. Zhu, Gnot: A general neural operator transformer for operator learning, inInterna- tional Conference on Machine Learning(PMLR, 2023) pp. 12556–12569

    work page 2023

  67. [67]

    N. B. Kovachki, S. Lanthaler, and A. M. Stuart, Op- erator learning: Algorithms and analysis, Handbook of Numerical Analysis25, 419 (2024)

    work page 2024

  68. [68]

    Malik, M

    M. Malik, M. O’Sullivan, B. Rodenburg, M. Mirhosseini, J. Leach, M. P. Lavery, M. J. Padgett, and R. W. Boyd, Influence of atmospheric turbulence on optical commu- nications using orbital angular momentum for encoding, Optics express20, 13195 (2012)

    work page 2012

  69. [69]

    S. Wang, Y. Cheng, Z. Yin, P. Durbin, and W. Li,ℓ 2/g2 hybrid RANS/LES model for simulating turbulent flows in the spectral element framework, Phys. Rev. Fluids10, 094902 (2025)

    work page 2025

  70. [70]

    Tamascelli, R

    D. Tamascelli, R. Rosenbach, and M. B. Plenio, Im- proved scaling of time-evolving block-decimation algo- rithm through reduced-rank randomized singular value decomposition, Physical Review E91, 063306 (2015)

    work page 2015