Pith. sign in

REVIEW 2 major objections 7 minor 71 references

The Spatial Complexity of Optical Computing and How to Reduce It

T0 review · 2 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Structural sparsity motivated by wave physics can shrink optical computing hardware to 1%–10% of conventional size while keeping accuracy near baseline.

desk verdict Worth engaging: the photonic-chip pruning results are solid and useful, but the headline 2D free-space thickness scaling law is not actually derived from the paper's own equations. read the letter →

arxiv 2411.10435 v2 pith:35LAFGFI submitted 2024-11-15 physics.optics cs.ETcs.LG

classification physics.opticscs.ETcs.LG
keywords spatialcomplexityopticalcomputingoverlappingnonlocalitystructuralsparsityneuralnetworksMach-Zehnderinterferometermeshfree-spaceopticsphotonicchips
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 asks how the physical size of an optical computer must grow as the mathematical operation it performs grows, and proposes a way to slow that growth. The central claim is that the right structural sparsity (local couplings for free-space optics and block-diagonal matrices for photonic chips) turns linear thickness scaling and quadratic component scaling into square-root and quasilinear scaling. Training neural networks under these constraints, the authors show that classification tasks like MNIST can be handled by devices roughly ten to one hundred times smaller, at the cost of only a few percent accuracy. If correct, this makes compact optical neural networks feasible and reframes the ultimate limit of optical computing as a trade-off between device size and accuracy.

What carries the argument

The load-bearing object is the overlapping nonlocality $C$ of a transverse cut: the number of communication cones that must cross from one side of the cut to the other. This number, through Eq. (1), lower-bounds the device thickness, and its maximum over all cuts is the quantity whose scaling with operation dimension $N$ the paper analyzes. The key identity, Eq. (2) and its derivation in the supplement (Eq. S12), gives $\mathbb{E}[\max(C)] = 2\max(d_{\parallel})(\sqrt{2N}-\max(d_{\parallel})) = O(N^{1/2})$ for local sparse kernels, versus $O(N)$ for dense, trivially sparse, or row sparse kernels. For photonic chips, the machinery is block-diagonalization: from a graph perspective, breaking an $N\times N$ kernel into roughly $N/N'$ blocks of size $N'\times N'$ requires about $(N/N')\cdot N'(N'-1)/2$ interferometers, which is quasilinear when the block size $N'$ is fixed.

What would settle it

Compute the thickness bound $\max(C/l_{\mathrm{cut}})$ for randomly generated local sparse matrices of increasing dimension $N$ with fixed $\max(d_{\parallel})$. If this quantity saturates or grows faster than $N^{1/2}$, the claimed scaling law is not supported; a direct check is to substitute Eq. S12 into Eq. (1) for the diagonal cut, which yields a thickness that stays roughly constant as $N$ grows.

Watch

Extended reading notes

Core claim

The paper introduces the spatial complexity of an optical computing system as the scaling of its physical footprint with the dimension $N$ of the kernel matrix that describes the input-output operation. It argues that ordinary sparsity does not reduce the overlapping nonlocality $C$ (the number of communication cones that must cross a transverse cut), which sets the thickness bound in Eq. (1); only structural sparsity does. For free-space optics, requiring all nonzero couplings to span only a limited in-plane distance (the local sparse form) makes $\max(C)$ scale as $O(N^{1/2})$ instead of $O(N)$, so the device thickness can grow as a square root of the operation dimension rather than linearly. For photonic chips, block-diagonal kernels reduce the number of required Mach-Zehnder interferometers from quadratic to quasilinear; a trained block-diagonal model with $10 \times 10$ blocks needs about 4,200 interferometers instead of roughly 300,000, with accuracy losses of a few percent on MNIST-like datasets. The authors present this as a design approach, space-efficient neuromorphic optics, in which neural pruning guided by wave physics dictates the structure of the optical hardware.

Load-bearing premise

The free-space square-root thickness law rests on the unverified assumption that large overlapping nonlocality values always occur at wide transverse apertures, so that the thickness bound $\max(C/l_{\mathrm{cut}})$ inherits the $O(N^{1/2})$ growth derived for $\max(C)$.

Editorial extensions

If this is right

  • Free-space optical neural networks can be made one to two orders of magnitude thinner for the same task, enabling ultrathin diffractive and metalens-based devices.
  • Block-diagonal photonic classifiers can cut the number of Mach-Zehnder interferometers by roughly 99% while keeping accuracy within a few percent of the unpruned network.
  • The same block-diagonal pruning extends to block-circulant weights and to edge-AI models such as MobileNetV2, suggesting practical hybrid optical-electronic accelerators.
  • Accuracy gains saturate as spatial resources grow, so the optimal device is the smallest structure that meets the accuracy target, not the largest one.
  • Optical systems can be made more space-efficient not only by changing wavelength or refractive index but by choosing or learning operation kernels with the right structural sparsity.

Reading between the lines

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

  • If the square-root thickness scaling holds for large $N$, free-space optical processors could scale much more slowly in footprint than their electronic counterparts, but the paper does not provide a general worst-case proof that large $C$ always aligns with wide apertures.
  • Block-diagonalization sacrifices inter-block connectivity, so tasks requiring long-range correlations or global mixing may degrade more than the MNIST-like benchmarks suggest.
  • Block-circulant structured weights appear to give slightly higher accuracy than block-diagonal weights at similar interferometer counts, hinting at a broader family of physics-friendly structured kernels beyond the two forms highlighted.
  • The diminishing-returns trend suggests a new design criterion for optical computing: choose the minimal spatial complexity that achieves a required accuracy, rather than maximizing raw performance.
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

2 major / 7 minor

Summary. The paper introduces a spatial-complexity framework for optical hardware, adapting D. A. B. Miller's overlapping-nonlocality (ONL) thickness bound to scaling laws in the matrix dimension N. For free-space optics, it analyzes trivial sparse, row sparse, and local sparse kernel matrices, derives scalings for E[max(C)], and claims that local sparse structure reduces required device thickness from O(N) to O(N^{1/2}); it then trains locally sparse optical neural networks with BIMT and reports thickness reductions of one to two orders of magnitude on MNIST-like benchmarks. For integrated photonics, it proposes block-diagonal and block-circulant weight matrices to reduce the number of Mach-Zehnder interferometers from quadratic to quasi-linear, with training experiments including a MobileNetV2 classifier. The paper includes public code and detailed supplementary derivations.

Significance. If the free-space scaling claim is properly established, the paper would provide a useful bridge between fundamental wave-physics bounds and approximate neuromorphic optical computing, and the block-diagonal photonic-chip result is likely correct and well demonstrated. Strengths include the self-contained random-matrix derivations, which are not circular, the empirical thickness and accuracy measurements with standard deviations, and the public code. However, the central 2D free-space thickness-scaling law currently conflates max(C) with max(C/lcut), so the main theoretical claim needs correction or an explicit proof before the paper can be accepted.

major comments (2)
  1. [Results: The Spatial Complexity of Free-Space Optics; Eq. (1)-(2); Methods: Calculating the Physical Thicknesses] Eq. (1) (second line) and the Methods define the 2D thickness bound in terms of max(C/lcut) over valid cuts, but Eq. (2) and Supplementary Eq. (S12) derive only E[max(C)] = O(max(d_parallel) sqrt(N)). The abstract and Fig. 1 caption's claim that local sparse kernels reduce the thickness scaling to O(N^{1/2}) therefore does not follow from the paper's equations. For an all-local LSmat with R = max(d_parallel), an interior cut of length lcut has C approximately 2R lcut, so C/lcut approximately 2R for such cuts, and corner cuts have lcut bounded below by the fixed peripheral mesh spacing used in the Methods; the thickness-relevant quantity appears to be O(1) for fixed R, which would be a stronger, not weaker, result. Please derive the scaling of max(C/lcut) explicitly and adjust the abstract, Eq. (2)'s interpretation, and Fig. 1 accordingly.
  2. [Fig. 3; Supplementary Note 1; Supplementary Note 2; Tables 1-3] The numerical scaling-law validation in Fig. 3 and Supplementary Note 1 sweeps over balanced cuts and reports E[max(C)], not the thickness-relevant E[max(C/lcut)]. The trained-ONN thickness reductions in Fig. 4 and Tables 1-3 are computed via max(C/lcut), but at fixed N=784, and Supplementary Note 2 explains the large reductions by an empirically observed allocation of large C to long cuts in LSONNs. This allocation is not proved for the local sparse matrix ensemble analyzed in Eq. (2)/(S12) or for the BIMT-trained subclass, leaving open the possibility that the favorable thickness scaling is a property of the trained models rather than of the structural sparsity class. Please provide either a direct bound on max(C/lcut) for local sparse matrices or numerical scaling data for max(C/lcut) as a function of N.
minor comments (7)
  1. [Eq. (1)] The printed form 'max(C) 1/lcut' is ambiguous and invites the conflation discussed above; write the bound as max over valid cuts of (C/lcut) and define lcut as the length of the specific cut achieving that maximum.
  2. [Fig. 3 caption] The vertical axis in Fig. 3 is max(C), not the thickness defined by Eq. (1); add a note that the 2D thickness is controlled by max(C/lcut) so that readers do not interpret these panels as thickness-scaling plots.
  3. [Eq. (2)] The third line of Eq. (2) should state the domain of validity, 0 <= max(d_parallel) <= sqrt(N/2), which currently appears only in the supplement; without this restriction the formula is not meaningful for large max(d_parallel).
  4. [Abstract and Fig. 2] The abstract's 'three-dimensional free-space optics' is inconsistent with the 1D/2D terminology used in Eq. (1) and Fig. 2; clarify that the physical structure is three-dimensional while the transverse planes are two-dimensional.
  5. [Supplementary Note 1, local sparse matrices] The statement that the diagonal cut maximizes the blue-shaded region, and hence max(C), is intuitive but not proven; provide a short argument or a lemma covering non-diagonal and non-balanced cuts.
  6. [Methods: Calculating the Physical Thicknesses] The normalization claim that each layer fits within a square of area 28^2 = 784 is not consistent with the stated hidden-layer spacing d_adj = 2.8 for a 10x10 grid, which gives a span of 25.2; reconcile the definition or specify that margins are included.
  7. [Fig. 5 caption] The caption lists numbers such as 'model b-d4 = 1084 requires ... = 3228' without identifying the first quantity as the number of free parameters and the second as the number of MZIs; label these explicitly.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the scaling-law derivations are self-contained matrix-ensemble calculations, and the author self-citations are background references rather than load-bearing support. The main weakness is a mathematical gap between max(C) and max(C/lcut), which is a correctness issue, not a circular reduction.

full rationale

The paper's load-bearing derivations do not reduce to their inputs by construction. The free-space scaling law (Eq. 2 / Eq. S12) is derived from the definition of local sparse matrices and a geometric count over a diagonal cut; it is an independent calculation, not an import of Miller's thickness bound. The photonic-chip block-diagonal result follows from a component count based on the Clements decomposition, an external and standard result, and the trained block-diagonal ONNs are evaluated on held-out MNIST-family data, making those results empirical rather than fitted predictions. The author self-citations (refs. 18, 27, 28, 49) are context or perspective references and are not used to justify the central construction's validity or uniqueness. One non-circular weakness should be noted for context: the paper presents the O(N^{1/2}) result as a 'thickness scaling law' (abstract and Fig. 1), but Eq. 1 and the Methods define thickness via max(C/lcut), not max(C); for the diagonal cut that maximizes C, lcut ~ sqrt(N), so the ratio need not grow as sqrt(N). This is a mismatch between the derived quantity and the claimed quantity, i.e., a mathematical gap rather than a circular derivation, and therefore does not raise the circularity score.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central theoretical results rest on Miller's thickness bound, a discrete port model, and the assumption that trained network weights can be directly translated into optical hardware. The reported reductions are conditioned on several hand-chosen hyperparameters such as neuron spacing, penalties, and pruning thresholds, though the scaling-law derivation itself has no fitted constants.

free parameters (5)
  • max(d_parallel), the locality radius R = variable; fixed at 6 port spacings in Fig. S7
    Sets which kernel entries are allowed. The derived scaling E[max(C)] = 2R(sqrt(2N)-R) is linear in R; the paper does not give a principled rule for choosing R in specific tasks.
  • Hidden/output layer neuron spacing d_adj = 2.8 for hidden layers, 1.0 for input
    Hand-chosen normalization in the thickness calculation. It rescales all reported thickness values, though comparisons among models use the same spacing.
  • BIMT locality penalty lambda_nl = 0.02 initially, scheduled to increase
    Controls how local the trained weights become. Different schedules would produce different thickness-accuracy trade-offs.
  • Off-block-diagonal penalty lambda_obd = not reported
    Used in Phase I of block-diagonal training (Eq. 6); the exact value is not given, which hampers exact reproduction.
  • Pruning thresholds tau and tau_neuron = tau in {0.01,...,0.2}, tau_neuron in {0.05,...,0.15}
    Swept to produce the trade-off curves. The headline size reductions are quoted at specific threshold pairs, so the reductions are conditional on these choices.
assumptions (4)
  • domain assumption Miller's thickness lower bound (Eq. 1) is valid for the considered transparent, non-absorbing optical systems.
    All absolute thickness numbers and the information-bottleneck interpretation are built on this bound from Ref. [17].
  • domain assumption Discretized port grids and valid-cut sweeps capture the continuous-field overlapping nonlocality.
    The paper computes C by counting discrete couplings between sampling points; the equivalence to the continuous physical quantity is assumed.
  • domain assumption Numerical scaling results obtained with balanced cuts only are representative of all valid cuts.
    The scaling-law simulations restrict the sweep to balanced cuts for speed; the paper does not fully prove that the maximum C always occurs on a balanced cut.
  • domain assumption Trained neural network weights map losslessly to the considered optical hardware, free-space layers or MZI meshes.
    The thickness and MZI counts are computed from digital weights; any overheads from physical implementation such as loss, crosstalk, or fabrication constraints are ignored.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Spatial Complexity of Optical Computing and How to Reduce It." pith.science (2026). https://pith.science/paper/35LAFGFI

@misc{pith2026241110435,
  author       = {Pith},
  title        = {Pith review of: The Spatial Complexity of Optical Computing and How to Reduce It},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/35LAFGFI}},
  note         = {Machine review of arXiv:2411.10435}
}
read the original abstract

Similar to algorithms, which consume time and memory to run, hardware requires resources to function. For devices processing physical waves, implementing operations needs sufficient "space," as dictated by wave physics. How much space is needed to perform a certain function is a fundamental question in optics, with recent research addressing it for given mathematical operations, but not for more general computing tasks, e.g., classification. Inspired by computational complexity theory, we study the "spatial complexity" of optical computing systems in terms of scaling laws - specifically, how their physical dimensions must scale as the dimension of the mathematical operation increases - and propose a new paradigm for designing optical computing systems: space-efficient neuromorphic optics, based on structural sparsity constraints and neural pruning methods motivated by wave physics (notably, the concept of "overlapping nonlocality"). On two mainstream platforms, free-space optics and on-chip integrated photonics, our methods demonstrate substantial size reductions (to 1%-10% the size of conventional designs) with minimal compromise on performance. Our theoretical and computational results reveal a trend of diminishing returns on accuracy as structure dimensions increase, providing a new perspective for interpreting and approaching the ultimate limits of optical computing - a balanced trade-off between device size and accuracy.

Figures

Figures reproduced from arXiv: 2411.10435 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

71 extracted references · 55 canonical work pages

  1. [1]

    Alan M. Turing. On computable numbers, with an application to the Entscheidungsproblem. Proceedings of the London Mathematical Society, 2(42):230–265, 1936

  2. [2]

    AI and Compute, May 2018

    Dario Amodei and Danny Hernandez. AI and Compute, May 2018. Accessed: 2024-09-17

  3. [3]

    Visual interpretability for deep learning: a survey.Frontiers of Information Technology & Electronic Engineering , 19(1):27–39, Jan 2018

    Quan-shi Zhang and Song-chun Zhu. Visual interpretability for deep learning: a survey.Frontiers of Information Technology & Electronic Engineering , 19(1):27–39, Jan 2018

  4. [4]

    Explainability for large language models: A survey

    Haiyan Zhao, Hanjie Chen, Fan Yang, Ninghao Liu, Huiqi Deng, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, and Mengnan Du. Explainability for large language models: A survey. ACM Trans. Intell. Syst. Technol. , 15(2), feb 2024

  5. [5]

    Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models, 2020. arXiv:2001.08361

  6. [6]

    Rae, Oriol Vinyals, and Laurent Sifre

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osindero, Karen Simonyan, Erich Elsen, Jack W. Rae, Oriol Vinyals, and Laurent Sifre...

  7. [7]

    Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus

    Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. Emergent abilities of large language models, 2022. arXiv:2206.07682

  8. [8]

    Physics of language models: Part 3.3, knowledge capacity scaling laws, 2024

    Zeyuan Allen-Zhu and Yuanzhi Li. Physics of language models: Part 3.3, knowledge capacity scaling laws, 2024. arXiv:2404.05405

Show all 71 references
  1. [9]

    The computational complexity of linear optics, 2010

    Scott Aaronson and Alex Arkhipov. The computational complexity of linear optics, 2010. arXiv:1011.3245

  2. [10]

    Ising formulations of many np problems

    Andrew Lucas. Ising formulations of many np problems. Frontiers in Physics , 2, 2014

  3. [11]

    Complex brain networks: graph theoretical analysis of structural and functional systems

    Ed Bullmore and Olaf Sporns. Complex brain networks: graph theoretical analysis of structural and functional systems. Nature Reviews Neuroscience, 10(3):186–198, Mar 2009

  4. [12]

    Kahng, Jens Lienig, Igor L

    Andrew B. Kahng, Jens Lienig, Igor L. Markov, and Jin Hu. VLSI Physical Design: From Graph Partitioning to Timing Closure. Springer Publishing Company, Incorporated, 1st edition, 2011

  5. [13]

    Xiaosheng Zhang, Kyungmok Kwon, Johannes Henriksson, Jianheng Luo, and Ming C. Wu. A large-scale microelectromechanical-systems-based silicon photonics lidar. Nature, 603(7900):253–258, Mar 2022

  6. [14]

    Full- colour 3d holographic augmented-reality displays with metasurface waveguides

    Manu Gopakumar, Gun-Yeal Lee, Suyeon Choi, Brian Chao, Yifan Peng, Jonghyun Kim, and Gordon Wetzstein. Full- colour 3d holographic augmented-reality displays with metasurface waveguides. Nature, 629(8013):791–797, May 2024

  7. [15]

    Zhou, Mark Harfouche, Colin L

    Kevin C. Zhou, Mark Harfouche, Colin L. Cooke, Jaehee Park, Pavan C. Konda, Lucas Kreiss, Kanghyun Kim, Joakim J¨ onsson, Thomas Doman, Paul Reamey, Veton Saliu, Clare B. Cook, Maxwell Zheng, John P. Bechtel, Aur´ elien B` egue, Matthew McCarroll, Jennifer Bagwell, Gregor Hors...

  8. [16]

    Igor L. Markov. Limits on fundamental limits to computation. Nature, 512(7513):147–154, Aug 2014

  9. [18]

    Toward ultrathin optics

    Francesco Monticone. Toward ultrathin optics. Science, 379(6627):30–31, 2023

  10. [19]

    Predicting parameters in deep learning

    Misha Denil, Babak Shakibi, Laurent Dinh, Marc’Aurelio Ranzato, and Nando de Freitas. Predicting parameters in deep learning. In Proceedings of the 26th International Conference on Neural Information Processing Systems - Volume 2 , NeurIPS’13, page 2148–2156, Red Hook, NY, USA...

  11. [20]

    Optimal brain damage

    Yann LeCun, John Denker, and Sara Solla. Optimal brain damage. In Advances in Neural Information Processing Systems , volume 2. Morgan-Kaufmann, 1989

  12. [21]

    Second order derivatives for network pruning: Optimal brain surgeon

    Babak Hassibi and David Stork. Second order derivatives for network pruning: Optimal brain surgeon. In Advances in Neural Information Processing Systems , volume 5. Morgan-Kaufmann, 1992

  13. [22]

    Learning both weights and connections for efficient neural network

    Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. In Advances in Neural Information Processing Systems , volume 28. Curran Associates, Inc., 2015

  14. [23]

    Pruning filters for efficient convnets

    Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017

  15. [24]

    Are sixteen heads really better than one? InAdvances in Neural Information Processing Systems, volume 32

    Paul Michel, Omer Levy, and Graham Neubig. Are sixteen heads really better than one? InAdvances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019

  16. [25]

    Reducing transformer depth on demand with structured dropout

    Angela Fan, Edouard Grave, and Armand Joulin. Reducing transformer depth on demand with structured dropout. In International Conference on Learning Representations , 2020

  17. [26]

    Sheared LLaMA: Accelerating language model pre-training via structured pruning

    Mengzhou Xia, Tianyu Gao, Zhiyuan Zeng, and Danqi Chen. Sheared LLaMA: Accelerating language model pre-training via structured pruning. In The Twelfth International Conference on Learning Representations , 2024

  18. [27]

    Nonlocal flat optics

    Kunal Shastri and Francesco Monticone. Nonlocal flat optics. Nature Photonics, 17(1):36–47, Jan 2023

  19. [28]

    Hyungki Shim, Francesco Monticone, and Owen D. Miller. Fundamental limits to the refractive index of transparent optical materials. Advanced Materials, 33(43):2103946, 2021

  20. [29]

    Seeing is believing: Brain-inspired modular training for mechanistic inter- pretability, 2023

    Ziming Liu, Eric Gan, and Max Tegmark. Seeing is believing: Brain-inspired modular training for mechanistic inter- pretability, 2023. arXiv:2305.08746

  21. [30]

    Spatial networks.Physics Reports, 499(1–3):1–101, February 2011

    Marc Barth´ elemy. Spatial networks.Physics Reports, 499(1–3):1–101, February 2011

  22. [31]

    MNIST handwritten digit database

    Yann LeCun, Corinna Cortes, and CJ Burges. MNIST handwritten digit database. ATT Labs [Online]. Available: http://yann.lecun.com/exdb/mnist, 2, 2010

  23. [32]

    Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms

    Han Xiao, Kashif Rasul, and Roland Vollgraf. Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms. CoRR, abs/1708.07747, 2017. 12

  24. [33]

    Deep learning for classical japanese literature

    Tarin Clanuwat, Mikel Bober-Irizar, Asanobu Kitamoto, Alex Lamb, Kazuaki Yamamoto, and David Ha. Deep learning for classical japanese literature. CoRR, abs/1812.01718, 2018

  25. [34]

    Prajit Ramachandran, Barret Zoph, and Quoc V. Le. Swish: a self-gated activation function, 2017. arXiv:1710.05941

  26. [35]

    Yardimci, Muhammed Veli, Yi Luo, Mona Jarrahi, and Aydogan Ozcan

    Xing Lin, Yair Rivenson, Nezih T. Yardimci, Muhammed Veli, Yi Luo, Mona Jarrahi, and Aydogan Ozcan. All-optical machine learning using diffractive deep neural networks. Science, 361(6406):1004–1008, 2018

  27. [36]

    Spatially varying nanophotonic neural networks

    Kaixuan Wei, Xiao Li, Johannes Froech, Praneeth Chakravarthula, James Whitehead, Ethan Tseng, Arka Majumdar, and Felix Heide. Spatially varying nanophotonic neural networks. Science Advances, 10(45):eadp0391, 2024

  28. [37]

    Clements, Peter C

    William R. Clements, Peter C. Humphreys, Benjamin J. Metcalf, W. Steven Kolthammer, and Ian A. Walmsley. Optimal design for universal multiport interferometers. Optica, 3(12):1460–1465, Dec 2016

  29. [38]

    Wim Bogaerts, Daniel P´ erez, Jos´ e Capmany, David A. B. Miller, Joyce Poon, Dirk Englund, Francesco Morichetti, and Andrea Melloni. Programmable photonic circuits. Nature, 586(7828):207–216, Oct 2020

  30. [39]

    Lithography-free reconfigurable integrated photonic processor

    Tianwei Wu, Marco Menarini, Zihe Gao, and Liang Feng. Lithography-free reconfigurable integrated photonic processor. Nature Photonics, 17(8):710–716, Aug 2023

  31. [40]

    R´ ıos Ocampo, and Mo Li

    Changming Wu, Haoqin Deng, Yi-Siou Huang, Heshan Yu, Ichiro Takeuchi, Carlos A. R´ ıos Ocampo, and Mo Li. Freeform direct-write and rewritable photonic integrated circuits in phase-change thin films. Science Advances, 10(1):eadk1361, 2024

  32. [41]

    Stein, Benjamin A

    Tatsuhiro Onodera, Martin M. Stein, Benjamin A. Ash, Mandar M. Sohoni, Melissa Bosch, Ryotatsu Yanagimoto, Marc Jankowski, Timothy P. McKenna, Tianyu Wang, Gennady Shvets, Maxim R. Shcherbakov, Logan G. Wright, and Pe- ter L. McMahon. Scaling on-chip photonic neural processors...

  33. [44]

    CIFAR-10 and CIFAR-100 datasets

    Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. CIFAR-10 and CIFAR-100 datasets. University of Toronto, 2009

  34. [45]

    cuSPARSE Library

    NVIDIA Corporation. cuSPARSE Library. https://developer.nvidia.com/cusparse

  35. [46]

    Horowitz, and William J

    Song Han, Xingyu Liu, Huizi Mao, Jing Pu, Ardavan Pedram, Mark A. Horowitz, and William J. Dally. EIE: efficient inference engine on compressed deep neural network. In Proceedings of the 43rd International Symposium on Computer Architecture, ISCA ’16, page 243–254. IEEE Press, 2016

  36. [47]

    Learning efficient convolutional networks through network slimming

    Zhuang Liu, Jianguo Li, Zhiqiang Shen, Gao Huang, Shoumeng Yan, and Changshui Zhang. Learning efficient convolutional networks through network slimming. In 2017 IEEE International Conference on Computer Vision (ICCV) , pages 2755– 2763, 2017

  37. [48]

    Peter L. McMahon. The physics of optical computing. Nature Reviews Physics , 5(12):717–734, Dec 2023

  38. [49]

    Exploring the role of metamaterials in achieving advantage in optical computing

    Yandong Li and Francesco Monticone. Exploring the role of metamaterials in achieving advantage in optical computing. Nature Computational Science , 4(8):545–548, Aug 2024

  39. [50]

    Wright, Peter F

    Alen Senanian, Logan G. Wright, Peter F. Wade, Hannah K. Doyle, and Peter L. McMahon. Programmable large-scale simulation of bosonic transport in optical synthetic frequency lattices. Nature Physics, 19(9):1333–1339, Sep 2023

  40. [51]

    Synthetic dimension in photonics

    Luqi Yuan, Qian Lin, Meng Xiao, and Shanhui Fan. Synthetic dimension in photonics. Optica, 5(11):1396–1405, Nov 2018

  41. [52]

    Gray, Ryoto Sekine, Luis Ledezma, Gordon H

    Robert M. Gray, Ryoto Sekine, Luis Ledezma, Gordon H. Y. Li, Selina Zhou, Arkadev Roy, Midya Parto, and Alireza Marandi. Large-scale time-multiplexed nanophotonic parametric oscillators, 2024. arXiv:2405.17355

  42. [53]

    Emergence of scaling in random networks.Science, 286(5439):509–512, 1999

    Albert-L´ aszl´ o Barab´ asi and R´ eka Albert. Emergence of scaling in random networks.Science, 286(5439):509–512, 1999

  43. [54]

    Strogatz

    Steven H. Strogatz. Exploring complex networks. Nature, 410(6825):268–276, Mar 2001

  44. [55]

    Small-world brain networks

    Danielle Smith Bassett and Ed Bullmore. Small-world brain networks. The Neuroscientist , 12(6):512–523, 2006. PMID: 17079517

  45. [56]

    Watts and Steven H

    Duncan J. Watts and Steven H. Strogatz. Collective dynamics of ‘small-world’ networks. Nature, 393(6684):440–442, Jun 1998

  46. [57]

    Efficient behavior of small-world networks

    Vito Latora and Massimo Marchiori. Efficient behavior of small-world networks. Phys. Rev. Lett. , 87:198701, Oct 2001

  47. [58]

    Graph structure of neural networks

    Jiaxuan You, Jure Leskovec, Kaiming He, and Saining Xie. Graph structure of neural networks. In Proceedings of the 37th International Conference on Machine Learning , ICML’20. JMLR.org, 2020

  48. [59]

    Light fields in complex media: Mesoscopic scattering meets wave control

    Stefan Rotter and Sylvain Gigan. Light fields in complex media: Mesoscopic scattering meets wave control. Rev. Mod. Phys., 89:015005, Mar 2017

  49. [60]

    in-plane

    Hang Chen. A pytorch implement of MobileNetV2 on cifar10. https://github.com/chenhang98/mobileNet-v2_cifar10, 2018. 13 METHODS Analyzing the Scaling Laws of the Three Types of Matrices The algorithms for generating sparse, row sparse, and local sparse matrices are detailed in ...

  50. [61]

    Definitions of the Three Types of Matrices 1 B

    Scaling Laws of Free-Space Optics 1 A. Definitions of the Three Types of Matrices 1 B. Numerical Method for Calculating max( C ) 2 C. Theoretical Analysis of the Scaling Laws 3 i. Introducing trivial sparsity in kernel matrix cannot lead to thinner optics 4 ii. How many free p...

  51. [62]

    Why can LSONN achieve a significant reduction in thickness? 8

  52. [63]

    Performance of NNs with Block-Circulant Weights 9

    ONNs with Block-Circulant Weights 8 A. Performance of NNs with Block-Circulant Weights 9

  53. [64]

    Algorithms for Constructing the Three Types of Structurally Sparse Matrices 11

  54. [65]

    extremely sparse

    Supplementary Dataset 12 References 16 1 Scaling Laws of Free-Space Optics In this section, we discuss the scaling laws, specifically how the maximum overlapping nonlocality , denoted as max(C), scales with the matrix dimension N for trivial sparse, row sparse, and local spars...

  55. [66]

    A balanced cut divides the input (output) plane into two parts, each containing an equal number ( N/2) of nodes

  56. [67]

    David A. B. Miller. Why optics needs thickness. Science, 379(6627):41–45, 2023

  57. [68]

    H. H. Zhu, J. Zou, H. Zhang, Y. Z. Shi, S. B. Luo, N. Wang, H. Cai, L. X. Wan, B. Wang, X. D. Jiang, J. Thompson, X. S. Luo, X. H. Zhou, L. M. Xiao, W. Huang, L. Patrick, M. Gu, L. C. Kwek, and A. Q. Liu. Space-efficient optical computing with an integrated chip diffractive ne...

  58. [69]

    De Mazancourt and D

    T. De Mazancourt and D. Gerlic. The inverse of a block-circulant matrix. IEEE Transactions on Antennas and Propagation, 31(5):808–810, 1983

  59. [70]

    Chen, and David Z

    Jiaqi Gu, Zheng Zhao, Chenghao Feng, Mingjie Liu, Ray T. Chen, and David Z. Pan. Towards area-efficient optical neural networks: An fft-based architecture. In 2020 25th Asia and South Pacific Design Automation Conference (ASP-DAC) , pages 476–481, 2020

  60. [71]

    Pan, and Ray T

    Shupeng Ning, Hanqing Zhu, Chenghao Feng, Christian Uselton, Jiaqi Gu, Rongxing Tang, David Z. Pan, and Ray T. Chen. Realization of a compact photoelectric platform for optical convolution processing. In CLEO 2024 , page SM4M.2. Optica Publishing Group, 2024

  61. [72]

    Soldano and E.C.M

    L.B. Soldano and E.C.M. Pennings. Optical multi-mode interference devices based on self-imaging: principles and applica- tions. Journal of Lightwave Technology , 13(4):615–627, 1995

  62. [73]

    Sometimes, multiple transverse apertures are associated with the same t

    Here, lcut refers to the length of the transverse aperture requiring the greatest thickness, defined as lcut≡ arg maxlcut(t) = arg maxlcut(C/lcut) over all valid cuts. Sometimes, multiple transverse apertures are associated with the same t. In such cases, we select the apertur...

  63. [74]

    Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen

    Mark Sandler, Andrew G. Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. MobileNetV2: Inverted residuals and linear bottlenecks. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018 , pages 4510–4...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.