Pith. sign in

REVIEW 5 major objections 6 minor 38 references

Quasi-Periodic Optical Key-Enabled Hybrid Cryptography: Merging Diffractive Physics and Deep Learning for High-Dimensional Security

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper reports a lensless optical cryptosystem in which a quasi-periodic phase mask encrypts a plaintext by diffraction, and a U-Net recovers the ciphertext phase from amplitude-only measurements, making decryption possible with…

desk verdict A plausible optical encryption demo with a real experiment, but the continuous-key-space claim is extrapolated from binary-phase, fixed-wavelength tests. read the letter →

arxiv 2505.23479 v1 pith:CXJ3TT7W submitted 2025-05-29 physics.optics physics.app-ph

Haiqi Gao , Yu Shao , Jiaming Liang , Xuehui Wang , Junren Wen , Yuchuan Shao , Yueguang Zhang , Weidong Shen
show 1 more author
Chenying Yang
This is my paper
classification physics.opticsphysics.app-ph
keywords opticalencryptionquasi-periodickeyphaseretrievalU-Netdiffractionmetasurfacespaceciphertextlossresilience
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

This paper proposes an optical encryption scheme in which a quasi-periodic phase mask, the Q-POK, encodes a plaintext image by diffraction, and a U-Net recovers the missing phase of the ciphertext from amplitude-only camera data plus the key parameters. The aim is to show that optical diffraction itself can carry the encryption load, shifting work from electronics to optics, while the mask's built-in periodic redundancy makes the physical key resilient to damage. If the scheme works as claimed, encryption keys become continuously tunable across wavelength, propagation distance, phase profile, and mask geometry, and decryption no longer requires full complex-field recording. The authors report that mismatched keys fail to decrypt, that a network trained on Digital-MNIST generalizes to E-MNIST, and that the ciphertext tolerates up to 20% random loss.

What carries the argument

The load-bearing object is the Quasi-Periodic Optical Key (Q-POK): a square phase plate assembled by repeating one random $n\times n$ phase tile $m\times m$ times, giving short-range randomness with long-range order. The encryption operation is convolution with the system's optical transfer function $H(d,\lambda)$ followed by pointwise multiplication with $K$; decryption is the same operation applied to the conjugated field, justified by time-reversal symmetry under the unitary-like composition of linear optical operators. The phase-retrieval engine is a U-Net, a convolutional encoder-decoder with skip connections, whose encoder receives ciphertext amplitude and key parameters and whose decoder outputs the ciphertext phase, trained with a periodic mean-squared-error loss that handles the $2\pi$ periodicity of phase. The Q-POK's periodic structure does double duty: it is the secret key, and its repeated units make the key statistically recoverable after damage.

What would settle it

Train the U-Net exactly as described at 671 nm with binary phase tiles, then encrypt a held-out plaintext with the same Q-POK at an unseen wavelength such as 532 nm and run the full decryption pipeline; if the recovered plaintext fails a predefined similarity threshold while ideal phase-conjugation decryption succeeds, the claimed continuous wavelength key dimension is falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a hierarchical mask--an $n\times n$ random phase tile repeated $m\times m$ times--acts as a physical key $K$ whose diffraction pattern encrypts the plaintext as $C = H(d,\lambda) \ast (P \odot K)$, and whose periodic redundancy lets the key be reconstructed from any surviving intact tile. Decryption exploits time-reversal symmetry: if the ciphertext's complex field is conjugated, modulated again by $K$, and propagated back, the plaintext is recovered. Because only the ciphertext intensity is recorded, the paper trains a U-Net that takes ciphertext amplitude and Q-POK parameters as input and predicts the ciphertext phase, using a periodic mean-squared-error loss that respects phase wrapping. The trained network decrypts images encrypted with different random Q-POK masks without retraining, generalizes across datasets, and the full loop is demonstrated experimentally with spatial light modulators and a 3D-printed input mask.

Load-bearing premise

The whole scheme rests on the U-Net, trained on simulated Digital-MNIST ciphertexts at one wavelength with binary phase tiles, predicting the ciphertext phase for the full continuous range of claimed key parameters; the paper only demonstrates a few random binary-phase geometries under otherwise fixed conditions.

Editorial extensions

If this is right

  • A single trained U-Net can decrypt images encrypted with different random Q-POK masks, so the training step does not need to be repeated for every key.
  • The physical key can be reconstructed from any remaining intact $n\times n$ unit, so scratched or fractured masks do not force re-enrollment of users.
  • Because only the ciphertext amplitude is stored or transmitted, data volume is roughly halved compared with recording the full complex field.
  • Wavelength, propagation distance, phase profile, and mask geometry act jointly as key parameters, making the key space continuous and high-dimensional.
  • The ciphertext can tolerate up to 20% random loss or corruption, so partial data loss during transmission does not prevent decryption.

Reading between the lines

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

  • Editorial inference: the continuous-key-space claim is only as strong as the U-Net's ability to interpolate to unseen key settings; a direct test would be decrypting ciphertexts made at a wavelength never seen in training, such as 532 nm instead of 671 nm.
  • Editorial inference: the periodic redundancy that enables key recovery is also a structural clue for an attacker who knows $m$ and $n$, so security rests mainly on secrecy of the random tile and of the continuous parameters, not on the mask being entirely unknown.
  • Editorial inference: the same quasi-periodic construction could be combined with polarization, incident angle, or orbital angular momentum, as the authors note; the key open test is whether the U-Net still recovers phase when two coupled key dimensions vary together.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The manuscript proposes an optical encryption/decryption system in which a quasi-periodic phase key (Q-POK) encrypts a complex optical field via diffraction, and a U-Net recovers the ciphertext phase from amplitude-only measurements; phase conjugation and re-propagation then recover the plaintext. The authors demonstrate the scheme in simulation for Digital-MNIST and E-MNIST and in a proof-of-concept experiment using spatial light modulators, and they claim a high-dimensional continuous key space spanning wavelength, propagation distance, phase modulation, and Q-POK geometry, along with robustness to key damage and ciphertext loss.

Significance. If the claims were fully supported, the approach would be a useful step toward hybrid optical cryptosystems: the Q-POK's periodic structure offers an unusual self-recovery property, the key-conditioned U-Net avoids per-key retraining in a limited tested regime, and the experiment links simulation to hardware. The paper is clearly structured and the concept of a compressible physical key is attractive. However, the headline security claim—a continuous high-dimensional key space—rests on untested generalization across the very parameters that define that space, and the quantitative evidence for cryptographic robustness is mostly generated by the same simulation model used to train the decoder. The experimental validation is qualitative and limited to fixed conditions.

major comments (5)
  1. [Results, Framework of Q-POK-based Hybrid Cryptography] The two central equations for encryption and decryption are typeset incorrectly and are mathematically ambiguous: the expression 'C = PH(d,λ)K=⊗' mixes convolution and Hadamard-product notation without defining operand order, and the decryption expression '( )* *,PC H d K λ= ⊗' is unreadable. A precise operator definition with correct conjugation and inverse propagation is required before the time-reversal argument can be assessed.
  2. [Methods, Time-Reversal Symmetry in Wave Propagation] The derivation assumes unitary-like operators T_i* T_i = I, but finite-aperture propagation, amplitude-only recording, and the experimental 3D-printed masks/SLMs do not satisfy this condition; the text itself calls this an 'ideal assumption.' Please provide quantitative bounds on the non-unitarity for the actual Fresnel numbers and apertures, or demonstrate that the residual error is small enough to support the claimed decryption fidelity.
  3. [Results, Multi-Dimensional Optical Key Design and Encryption Security Analysis] The central claim of a continuously tunable key space (wavelength, distance, phase, geometry) is not demonstrated. Every successful decryption in Figs. 3 and 5 uses λ = 671 nm, binary phase (0, π), and one of two geometries; the only cross-key test, Fig. 3e, varies the random unit realization while holding all other parameters fixed. There is no experiment or simulation for unseen wavelengths, continuous phase profiles, or distance values beyond the three distances used in the security statistics (d1, d2, d3). Either provide generalization tests over the claimed key dimensions or revise the claim to the tested discrete parameter set.
  4. [Results, Encryption Security Analysis; Methods, Training details] The security evaluation and the U-Net training share the same forward diffraction model: ciphertexts are generated by simulating the same H(d,λ) and Q-POK that are used as training data. Consequently, the reported phase-retrieval success and statistical metrics (entropy, histogram variance, cross-correlation, PCA) are largely a self-consistency check. An independent validation—for example, a hold-out set generated with a different diffraction solver or measured experimentally—is needed to support the security claims.
  5. [Results, Experimental validation] The experimental decryption results in Fig. 5 are only assessed qualitatively ('clearly distinguishable'). Given that the experiment is the only independent check of the whole pipeline, please report quantitative fidelity metrics (e.g., SSIM, MSE, or classification accuracy) for the decrypted images, for both the ideal-conjugation and U-Net routes, and state how many patterns were tested.
minor comments (6)
  1. [Abstract and Results, Encryption Security Analysis] The abstract states 'reducing inter-class distances by over 50%,' but the Results section reports a decrease of average Euclidean distance from over 38,000 to approximately 13,000; please align the wording and define which distance is being reported.
  2. [Methods, Training details of the U-net] The training details do not specify how the U-Net is conditioned on d and λ; the text says it receives 'Q-POK modulation parameters,' but the architecture description in Fig. 1(c) only shows concatenation of amplitude and Q-POK. Clarify the input encoding.
  3. [Figure 1 and Figure 3] Figure 1(b) labels contain typos ('Eva l on', 'EMNIST') and Fig. 3(e) contains 'Trianed'; please proofread the figures.
  4. [Methods, PMSE loss] The PMSE loss equation in Methods is typeset incorrectly and is unreadable; please provide a clean definition.
  5. [Results, Framework] The notation 'ciphertext*' is used without a formal definition; define it as the complex conjugate of the ciphertext field.
  6. [Discussion] The Discussion states the current implementation combines only d and λ, but the Results claims phase modulation and geometry as part of the key space; reconcile these statements.

Circularity Check

2 steps flagged · score 4.0 of 10

In-simulation phase retrieval fits the same forward diffraction model that defines the encryption, while the 'unprecedented' continuous key-space claim is extrapolated from a single-wavelength, binary-phase experiment; the SLM validation independently grounds only the narrow tested configuration.

  1. fitted input called prediction [Results, 'Framework of Q-POK-based Hybrid Cryptography' (Eq. 1 and U-Net paragraph); Methods, 'Training details of the U-net for ciphertext* restruction']
    "The U-Net receives the amplitude of the ciphertext and all the Q-POK modulation parameters, and it outputs the phase of the ciphertext required for the decryption process. ... The target of the U-Net is −φ, where the φ is the phase term of the optical field of ciphertext. During the training process, we employ a periodic mean squared error (PMSE) loss function... to optimize model performance."

    The network's training targets (ciphertext phases) and evaluation inputs (ciphertext amplitudes) are both generated by the same forward model C = H(d,λ)*(P⊗K) used for encryption, with ground-truth phase computed from that model. In-simulation decryption successes (Fig. 3, E-MNIST, S2 robustness, simulated columns of Fig. 5) therefore measure how well the network inverted the simulator on the simulator's own output distribution — statistically forced fitting, not an independent derivation of the decryption capability. The only out-of-simulator evidence, the SLM experiment (Fig. 5b), uses a single narrow configuration (0/π binary phase, m=8, n=7, 671 nm).

  2. self definitional [Results, 'Encryption Security Analysis' and Fig. 3 caption]
    "Notably, even with complete knowledge of all other system parameters, including diffraction distances (d) and wavelength (λ), successful decryption was achieved only when the Q-POK configuration precisely identical to the one used during encryption was applied, as shown in Fig. 3 a-d. This observation strongly confirms the robustness and security of the proposed Q-POK design."

    The mismatched-key failure is an algebraic consequence of the construction: the decryption expression P = H*(d,λ)⊗(C*⊗K) inverts C = H(d,λ)*(P⊗K) only when the same K is applied, so 'wrong key ⇒ no decryption' holds for any keyed invertible transform by definition. The paper presents this definitional property as strong confirmation of 'robustness and security,' so the validation reduces to the scheme's own equations rather than to an independent security analysis. This does not question the correctness of the underlying wave physics, but the security inference is entailed by the formalism it claims to test.

full rationale

The core physics — Fresnel diffraction and phase-conjugation time-reversal — is standard, externally established material, and the Methods section derives the decryption identity self-containedly from unitary operators (T_i* T_i = I ⇒ E'_out = E_0), so the decryption math is not circular. The circularity pressure concentrates on the learned phase retrieval. The U-Net is trained on ciphertexts produced by the same forward model C = H(d,λ)*(P⊗K) that defines the encryption, with ground-truth phases computed from that model; every simulated decryption success (Fig. 3 cross-key tests, E-MNIST, S2 loss robustness, simulated columns of Fig. 5) is therefore an in-simulator result. A network trained to invert the simulator's mapping will reproduce phases on the simulator's output distribution; this is fitting, not independent derivation. What breaks the circle is the physical experiment: a simulation-trained U-Net decrypted real SLM-generated ciphertexts (Fig. 5b), a genuine out-of-distribution check of the basic functionality. However, that experiment, like all demonstrations, uses a single wavelength (671 nm), binary 0/π phase, and two geometries; the claim of an unprecedented security dimensionality (m×n×d×λ×φ) with continuously tunable keys is an extrapolation beyond every demonstrated configuration, and the paper's own Discussion limits the current implementation to d and λ among the listed dimensions. Self-citations (refs 34, 38) support only peripheral implementation alternatives, and the phase-conjugation principle is derived in-house rather than merely cited, so no load-bearing citation chain exists. Net result: partial circularity — the headline key-space claim is neither derived nor demonstrated, while the matched-key decryption mechanism itself is experimentally anchored.

Assumptions & free parameters 6 free parameters · 3 assumptions · 1 invented entities

The system rests on standard scalar diffraction and unitary propagation assumptions, plus the empirical assumption that a trained U-Net can invert the forward model across the claimed key space. The main hand-chosen parameters are the Q-POK geometry, propagation distance, wavelength, and binary phase states. The Q-POK itself is an engineered invention with in-paper evidence.

free parameters (6)
  • Q-POK unit size n = 7 or 8
    Chosen by hand; the random phase plate size defines the key granularity and is central to the architecture in Figure 2a.
  • Q-POK replication count m = 8 or 7
    Chosen by hand; forms the m by m periodic array and determines the damage-recovery property.
  • Propagation distance d = 1400, 2800, 4200 mm in simulation; 400/1000/1000/400 mm in experiment
    Continuously tunable key parameter; the system is evaluated only at discrete values, not continuously.
  • Wavelength lambda = 671 nm
    Fixed in both experiment and metasurface simulation; used as a key parameter but never varied in the tests.
  • Phase modulation states = 0 and pi (binary)
    All experiments restrict phase to binary values; continuous phase is claimed as a key dimension but is not tested.
  • U-Net training hyperparameters = 20 epochs, learning rate 0.001 with cosine annealing, architecture channels per Figure 1c
    Selected without ablation; these choices affect decryption fidelity and the claimed cross-key generalization.
assumptions (3)
  • domain assumption Optical propagation operators are unitary: T_i^* T_i = I for each linear operator (diffraction, phase modulation, amplitude modulation)
    Methods section, Time-Reversal Symmetry. This idealizes lossless, infinite-aperture propagation, which is not strictly true in the finite SLM experiment.
  • domain assumption Scalar Fresnel diffraction with a single OTF H(d, lambda) accurately models the encryption path
    Results section, encryption equation. Ignores polarization, aberrations, and alignment errors; the paper partially compensates with a Fresnel phase correction but does not model residual errors.
  • ad hoc to paper A U-Net trained on simulated Digital-MNIST ciphertexts generalizes to unseen keys, unseen patterns (E-MNIST), and real experimental hardware
    Results section, Figures 3e and 1b. The key-conditioned input is intended to handle new keys, but only a few configurations are tested.
invented entities (1)
  • Quasi-Periodic Optical Key (Q-POK) independent evidence
    purpose: Physical encryption layer made of an n by n random phase unit replicated m by m times; provides the key space, damage recovery, and a phase prior for the U-Net
    The structure is demonstrated with SLM experiments and metasurface simulations in the paper. A falsifiable handle exists because the phase profile can be measured directly, so it is not an unfalsifiable entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quasi-Periodic Optical Key-Enabled Hybrid Cryptography: Merging Diffractive Physics and Deep Learning for High-Dimensional Security." pith.science (2026). https://pith.science/paper/CXJ3TT7W

@misc{pith2026250523479,
  author       = {Pith},
  title        = {Pith review of: Quasi-Periodic Optical Key-Enabled Hybrid Cryptography: Merging Diffractive Physics and Deep Learning for High-Dimensional Security},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CXJ3TT7W}},
  note         = {Machine review of arXiv:2505.23479}
}
read the original abstract

Optical encryption inherently provides strong security advantages, with hybrid optoelectronic systems offering additional degrees of freedom by integrating optical and algorithmic domains. However, existing optical encryption schemes heavily rely on electronic computation, limiting overall efficiency, while the physical keys are susceptible to damage, compromising both security and system stability. To overcome these challenges, we introduce the Quasi Periodic Optical Key (QPOK), which combines long range order with short range disorder, enabling enhanced security and robustness against damage within a single platform. By leveraging diffraction symmetry, our design enables optics-driven encryption, effectively shifting the optoelectronic balance toward photonic processing. Moreover, we innovatively apply deep learning to reconstruct the complex optical ciphertext field using only amplitude data and cryptographic keys, simultaneously achieving data compression and improved security. Within this framework, the key space includes continuously tunable parameters such as wavelength, propagation distance, phase modulation, and Q-POK geometry, significantly expanding cryptographic diversity. Our system also demonstrates robust cryptographic reliability by reducing inter-class distances by over 50% and tolerating up to 20% ciphertext loss. Our framework represents a new generation of physically grounded, algorithmically enhanced optical cryptosystems, laying a foundational pathway for scalable, hardware-integrated information security paradigms.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 36 canonical work pages

  1. [1]

    & Sheridan, J

    Liu, S., Guo, C. & Sheridan, J. T. A review of optical image encryption techniques. Optics & Laser Technology 57, 327–342 (2014)

    work page 2014

  2. [2]

    A review of single and multiple optical image encryption techniques

    Hazer, Abdurrahman, and Remzi Yıldırım. A review of single and multiple optical image encryption techniques. Journal of Optics 23.11, 113501 (2021)

    work page 2021

  3. [3]

    9-bit spin-and wavelength-encoded hologram using a hybrid metasurface

    YUAN, Huan; ZHONG, Zheqiang; ZHANG, Bin. 9-bit spin-and wavelength-encoded hologram using a hybrid metasurface. Optics Letters, 49.3: 498-501 (2024)

    work page 2024

  4. [4]

    Wu, X. et al. Enhanced Reconfigurable Visual Cryptography Strategies Utilizing Optical Metasurfaces. ACS Appl. Mater. Interfaces 17, 11300–11308 (2025)

    work page 2025

  5. [5]

    & Povinelli, M

    Audhkhasi, R. & Povinelli, M. L. Full Spectral Image Encryption in the Infrared using an Electrically Reconfigurable Metasurface and a Matched Detector. Advanced Photonics Research 5, 2300254 (2024)

    work page 2024

  6. [6]

    Georgi, P. et al. Optical secret sharing with cascaded metasurface holography. Sci. Adv. 7, eabf9718 (2021)

    work page 2021

  7. [7]

    Zhang, F. et al. Meta-optics empowered vector visual cryptography for high security and rapid decryption. Nat Commun 14, 1946 (2023)

    work page 2023

  8. [8]

    Zheng, P. et al. Metasurface-based key for computational imaging encryption. Sci. Adv. 7, eabg0363 (2021)

    2021

Show all 38 references
  1. [9]

    Huang, K. et al. Silicon multi‐meta‐holograms for the broadband visible light. Laser & Photonics Reviews 10, 500–509 (2016)

    2016

  2. [10]

    Zhu, Y . et al. Optical information hiding based on speckle encoding with dual- multiplexing interferometry. Optics Communications 573, 131022 (2024)

    2024

  3. [11]

    & Chen, W

    Zhou, L., Xiao, Y . & Chen, W. Learning complex scattering media for optical encryption. Opt. Lett. 45, 5279 (2020)

    2020

  4. [12]

    Applications of optical logic- operated moiré in moiré topography and deflectometry

    Zhang, Jiajun, et al. Applications of optical logic- operated moiré in moiré topography and deflectometry. Applied optics 31.34, 7355-7360 (1992)

    1992

  5. [13]

    Yu, Z. et al. High-security learning-based optical encryption assisted by disordered metasurface. Nat Commun 15, 2607 (2024)

    2024

  6. [14]

    Tan, X. et al. Secure optical memory system with polarization encryption. Appl. Opt. 40, 2310 (2001)

    2001

  7. [15]

    Su, Y . et al. Optical image conversion and encryption based on structured light illumination and a diffractive neural network. Appl. Opt. 62, 6131 (2023)

    2023

  8. [16]

    Optical multiple- image encryption in diffractive- imaging-based scheme using spectral fusion and nonlinear operation

    Qin, Yi, et al. Optical multiple- image encryption in diffractive- imaging-based scheme using spectral fusion and nonlinear operation. Optics express 24.23, 26877-26886 (2016)

    2016

  9. [17]

    Peng, D. et al. Optical coherence encryption with structured random light. PhotoniX 2, 6 (2021)

    2021

  10. [18]

    Ni, R., Wang, F., Wang, J. & Hu, Y . Multi-Image Encryption Based on Compressed Sensing and Deep Learning in Optical Gyrator Domain. IEEE Photonics Journal 13, (2021)

    2021

  11. [19]

    & Javidi, B

    Matoba, O. & Javidi, B. Secure holographic memory by double-random polarization encryption. Appl. Opt. 43, 2915 (2004)

    2004

  12. [20]

    & Horisaki, R

    Mashiko, R., Naruse, M. & Horisaki, R. Diffraction casting. Adv. Photon. 6, (2024)

    2024

  13. [21]

    Liu, Z. et al. Color image encryption by using Arnold transform and color -blend operation in discrete cosine transform domains. Optics Communications 284, 123–128 (2011)

    2011

  14. [22]

    Optical image encryption based on XOR operations

    Han, JongWook, et al. Optical image encryption based on XOR operations. Optical Engineering 38.1, 47-54 (1999)

    1999

  15. [23]

    Large- scale scattering -augmented optical encryption

    Bian, Liheng, et al. Large- scale scattering -augmented optical encryption. Nature Communications 15.1, 9807 (2024)

    2024

  16. [24]

    Bai, B. et al. Data‐Class‐Specific All‐Optical Transformations and Encryption. Advanced Materials 35, 2212091 (2023)

    2023

  17. [25]

    Abuturab, M. R. Color information security system using Arnold transform and double structured phase encoding in gyrator transform domain. (2013)

    2013

  18. [26]

    Zhou, T. et al. Large-scale neuromorphic optoelectronic computing with a reconfigurable diffractive processing unit. Nat. Photonics 15, 367–373 (2021)

    2021

  19. [27]

    Zhou, T. et al. In situ optical backpropagation training of diffractive optical neural networks. Photon. Res. 8, 940 (2020)

    2020

  20. [28]

    & Fang, L

    Yuan, X., Wang, Y ., Xu, Z., Zhou, T. & Fang, L. Training large -scale optoelectronic neural networks with dual-neuron optical-artificial learning. Nat Commun 14, 7110 (2023)

    2023

  21. [29]

    Yan, T. et al. Fourier-space Diffractive Deep Neural Network. PHYSICAL REVIEW LETTERS (2019)

    2019

  22. [30]

    & Bai, X

    Xu, X., Guo, S., Chen, J. & Bai, X. Pyramid -ladder diffractive neural network for visual recognition. Optics & Laser Technology 176, 110937 (2024)

    2024

  23. [31]

    Ryou, A. et al. Free-space optical neural network based on thermal atomic nonlinearity. Photon. Res. 9, B128 (2021)

    2021

  24. [32]

    Time ‐lapse image classification using a diffractive neural network

    Rahman, Md Sadman Sakib, and Aydogan Ozcan. Time ‐lapse image classification using a diffractive neural network. Advanced Intelligent Systems 5.5, 2200387 (2023)

    2023

  25. [33]

    Rahman, M. S. S., Li, J., Mengu, D., Rivenson, Y . & Ozcan, A. Ensemble learning of diffractive optical networks. Light Sci Appl 10, 14 (2021)

    2021

  26. [34]

    All-optical combinational logical units featuring fifth-order cascade

    Gao, Haiqi, et al. All-optical combinational logical units featuring fifth-order cascade. Chip 3.4, 100112 (2024)

    2024

  27. [35]

    Shen, C. -Y. et al. All-optical phase conjugation using diffractive wavefront processing. Nat Commun 15, 4989 (2024)

    2024

  28. [36]

    Li, J. et al. All-optical complex field imaging using diffractive processors. Light Sci Appl 13, 120 (2024)

    2024

  29. [37]

    Xue, Z. et al. Fully forward mode training for optical neural networks. Nature 632, 280–286 (2024)

    2024

  30. [38]

    Relief-Surface-Based On-Chip Hybrid Diffraction Neural Network Enabled by Authentic All-Optical Fully Connected Architecture

    Gao, Haiqi, et al. Relief-Surface-Based On-Chip Hybrid Diffraction Neural Network Enabled by Authentic All-Optical Fully Connected Architecture. ACS Photonics 11.11, 4818-4829 (2024)

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.