Pith. sign in

REVIEW 3 major objections 5 minor 31 references

Iterative optimization of photonic crystal nanocavity designs by using deep neural networks

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read An iterative deep-learning loop finds an L3 photonic-crystal cavity with Q = 11 million, more than twice the previous record.

desk verdict A credible iterative deep-learning optimization of L3 cavities with an 11M-Q FDTD result, but the record claim needs convergence evidence before it fully lands. read the letter →

arxiv 1908.03702 v1 pith:VU2PQI23 submitted 2019-08-10 physics.comp-ph physics.optics

Takashi Asano , Susumu Noda This is my paper · ORCID
classification physics.comp-phphysics.optics PACS 42.70.Qs42.60.Da
keywords photoniccrystalnanocavityqualityfactordeepneuralnetworkiterativeoptimizationL3cavityfinite-differencetime-domaingradient-basedsearchhigh-dimensionaldesignspace
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

The paper proposes a closed-loop recipe for discovering photonic-crystal nanocavity designs with record-high quality factors: train a neural network to predict Q from air-hole displacements, use the network's gradient to propose new structures, compute their Q with 3D finite-difference time-domain (FDTD) simulation, add those to the training set, and repeat. Applied to a standard silicon L3 cavity with 25 independent displacement parameters, the loop finds a design with $Q = 1.10\times 10^{7}$ within 101 iterations and 8,070 total simulated samples, more than twice the best Q previously reported for this cavity type. The key ingredient is a search penalty that pushes candidate structures away from already-evaluated designs, which the paper shows yields better results than staying near the current best or adding random exploration. A sympathetic reader should care because the method automates the exploration of the high-dimensional hole-position space that manual, physics-guided optimization cannot cover.

What carries the argument

The load-bearing mechanism is the iterate: neural-network regression plus a gradient search whose loss function contains artificial penalty terms. The network maps the 25-dimensional hole-displacement vector to $\log_{10}Q$ and provides a smooth surrogate whose gradient does not vanish in high dimensions. Search is performed by minimizing $L' = |\log_{10}Q_{\text{target}} - \log_{10}Q_{\text{NN}}|^2 + \text{artificial loss}$, with $Q_{\text{target}}=10^{8}$, starting from random initial structures. Three artificial losses are tested: (A) squared distance to the best previous structure, (B) squared distance to a random initial structure, and (C) a sum of inverse distances to all dataset structures; (C) makes the search avoid already-known regions. The paper's demonstration that strategy (A+C) finds the highest Q shows that forcing exploration of unexplored parameter space is what carries the optimization to record values.

What would settle it

Recalculate the Q of the paper's best cavity (Figure 5c) with an independent FDTD solver on a sequence of finer meshes; if the extrapolated Q does not exceed 5.3 million, the claimed improvement is not real. A second decisive check is to rerun the 101-round loop from the same initial dataset and confirm that the $Q=1.10\times 10^{7}$ structure appears within 8,070 Q evaluations.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that iteratively augmenting the training data with the structures the network itself proposes turns a rough regression model into a practical search engine for very high Q cavities. Starting with 1,000 random L3 structures (Q spread from $10^{3}$ to $10^{5}$), ten networks are trained to predict $\log_{10}Q$ from the 25 independent in-plane hole displacements (the 50-hole optimization region reduced by mirror symmetry). Each round, 70 candidates are produced by gradient descent on a loss that targets $Q=10^{8}$ while one of three artificial penalty terms constrains the explored region, and their true 3D-FDTD Q values are appended. After 101 rounds, the best structures found are $Q=5.75\times 10^{6}$, $Q=9.12\times 10^{6}$, and $Q=1.10\times 10^{7}$ for strategies (A), (A+B), and (A+C), respectively. The $Q=1.10\times 10^{7}$ result is presented as more than twice the previous records of $4.2\times 10^{6}$ (genetic algorithm) and $5.3\times 10^{6}$ (leaky-mode visualization) for an Si L3 cavity, with a comparable number of FDTD samples.

Load-bearing premise

The central premise is that the simulated Q factors from 3D-FDTD are accurate enough to rank cavities at Q near 10 million, and that the authors' FDTD conditions match those used for the earlier record values; neither is verified by an error or convergence test.

Editorial extensions

If this is right

  • The loop should transfer to other cavity geometries and other figures of merit, since the network output and the validation simulator can be changed independently.
  • The 8,070-sample dataset accumulated under strategy (A+C) is a reusable resource; later optimization runs can warm-start from it instead of sampling randomly.
  • The result sets a new benchmark for L3 nanocavity design: a method should beat $Q = 1.10\times 10^{7}$ with no more FDTD evaluations to claim an improvement.
  • The linear training cost of the neural network versus the cubic cost of Bayesian optimization makes this kind of iterative surrogate loop the more scalable option for high-dimensional design spaces.

Reading between the lines

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

  • A natural extension the authors leave implicit is applying repulsive exploration (C) to multi-objective design, where the network predicts several metrics and the same 'avoid known points' pressure maps trade-offs between Q and modal volume.
  • The inverse-distance penalty is a crude stand-in for prediction uncertainty; a testable variant would replace it with the variance across the ten networks and use that variance to guide exploration.
  • If the FDTD record survives an independent mesh-converged check, the same loop should be carried through fabrication; a measured Q near $10^{7}$ would turn the theoretical record into a device claim.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes an iterative optimization method for photonic-crystal nanocavity designs. A deep neural network is trained on a dataset of random hole-displacement structures and their FDTD-computed Q factors; candidate structures are then generated by gradient search on the network's predicted Q, subject to artificial loss terms that constrain exploration. The candidates are validated by 3D-FDTD and appended to the training set for the next iteration. The method is demonstrated on a silicon L3 cavity with 25 free parameters (after enforcing mirror symmetries), using 101 iterations and 8070 total FDTD evaluations. The reported best design has Q = 1.10 × 10^7, obtained with their 'A+C' exploration strategy, which the authors claim is more than twice the previous record L3-cavity Q values from a genetic algorithm [16] and from leaky-mode visualization [17]. The paper also compares the approach with Bayesian optimization and argues that the gradient-based search over a NN surrogate is better suited to high-dimensional parameter spaces.

Significance. If the numerical Q values are reliable, this is a valuable contribution: the method is not circular in the sense that the final Q is computed by FDTD rather than predicted by the network, and the iterative accumulation of validated samples is a sound and easily reproduced idea. The reported comparison across three exploration strategies and the discussion of Bayesian optimization are useful. The central quantitative claim, however, rests on two assumptions that are not verified in the manuscript: that the 3D-FDTD results are converged and accurate at the level of a factor of two, and that the earlier literature Q values [16,17] are directly comparable to the authors' own simulator settings. Because these assumptions are load-bearing for the headline record claim, the paper needs additional numerical evidence before the claim can be accepted.

major comments (3)
  1. [§3.3, §3.4, Fig. 3] The central result, Q = 1.10 × 10^7, is presented without any convergence or error analysis for the 3D-FDTD calculation. For Q values above 10^7, small numerical leakage due to finite grid resolution, PML placement, or simulation domain size can change the computed Q by a large fraction, and a factor-of-two claim is not robust to such effects. The authors should report a mesh-convergence study (e.g., Q vs. grid size for the final structure), a test of PML distance and boundary conditions, and an estimate of the numerical uncertainty of the reported Q values. Without this, the value Q = 1.10 × 10^7, and its comparison with previous records, is not quantitatively established.
  2. [§4.2, §3.1] The comparison with the previous record values from [16] and [17] assumes that the authors' FDTD settings produce Q values directly comparable to those in the cited references. The text itself notes that the slab thickness differs (0.5366a here versus 0.55a in [16]), and the hole radius and other simulation conditions are said to be 'the same as in [19]' without giving details. If the reference values were computed with different mesh sizes, domain sizes, or PML parameters, the claimed factors of 2.6 and 2.1 could be artifacts of numerical settings. The authors should either recompute the reference structures with their own FDTD settings or otherwise demonstrate that the cited Q values are directly comparable.
  3. [§3.4, Fig. 5] The final optimized structure is shown only as a displacement-vector plot; the numerical coordinates of the air holes in the record Q = 1.10 × 10^7 design are not provided. Since the paper's purpose is to demonstrate a useful design tool and a quantitative record, the actual optimized geometry should be made available, either as a table of the 25 displacement parameters or as a data file. This is needed for independent verification and for other researchers to use the result.
minor comments (5)
  1. [Headings] The manuscript contains two sections numbered '3.2', one titled 'Learning phase' and the other 'Structure search phase'; the second should be renumbered.
  2. [Eq. (1)] Equation (1) shows 'log 0' where it should read 'log10' in the teacher-data term; this appears to be a typesetting error that should be corrected.
  3. [Eq. (5)] Equation (5) contains a stray '≤ .' and the displayed formula could be misinterpreted as an inequality rather than a penalty term; please restate it as a regular sum of inverse distances.
  4. [§3.2] The claim that the 10 neural networks 'learn the dataset in different orders' is not specified further; a sentence describing how the ordering is randomized would improve reproducibility.
  5. [§4.1] The phrase 'computation costs for first principles calculations increased only by 8 times (from 1070 to 8070 sample cavities)' is unclear because the first round uses 1000 initial plus 70 candidates, and later rounds add 70 per round; the count should be made explicit.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the reported Q factors are outputs of 3D-FDTD, not of the neural network surrogate.

full rationale

The paper's optimization loop is a standard active-learning/surrogate-optimization cycle: random structures are evaluated by 3D-FDTD, a neural network is trained to approximate the Q factor, the trained network proposes candidate structures via gradient search with artificial losses, and those candidates are then re-evaluated by 3D-FDTD before being added to the training set. The final reported Q values (5.75e6, 9.12e6, 1.10e7) are FDTD-computed validation values, not outputs of the neural network. The paper is explicit about this separation: in Section 3.4 it states that the regression function 'is only employed to identify structures with potentially higher Q factors (via the gradient method)' and that the corresponding QNN values 'are not discussed in the following.' Thus there is no fitted parameter or surrogate prediction being renamed as a measured result. The neural network is a proposal generator whose proposals are independently checked by FDTD, so the central claim does not reduce to the network's own predictions. Self-citations to the authors' earlier work [19] provide the network architecture, hyperparameters, and FDTD calculation conditions; these are implementation details, not load-bearing premises for the conclusion. The comparison with previous L3 cavity records from Refs. [16] and [17] is an external benchmark comparison and not an input to the optimization. The absence of an FDTD mesh-convergence or boundary-convergence study is a legitimate correctness and reproducibility concern, but it is not a circularity issue. No circular reasoning or definitional equivalence between inputs and outputs was found.

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

The central claim rests on the accuracy of FDTD Q calculations, the symmetry-restricted parameter space, and the NN's ability to guide search. No physical constant is fitted; the listed hyperparameters are standard ML choices and are not tuned to the final Q. The method does not introduce new physical entities.

free parameters (4)
  • lambda (weight decay coefficient) = 0.00333
    Determined by 10-fold cross-validation on the initial 1000-sample dataset (Eq. 1). It is a regularization hyperparameter, not tuned to the final Q result.
  • lambda prime values for artificial losses = e.g., 3, 1, 0.1, 0.01, 0.001, 0.0001, 0.00001 for strategy (A)
    Chosen by hand to generate diverse candidate structures (Eqs. 3-5). Not fitted to data; the final Q is determined by FDTD, not by these values.
  • Air hole displacement range = ±0.1a
    Chosen based on prior manual optimizations of L3 cavities [2,17] and assumed to contain high-Q designs; restricts the search space.
  • NN training hyperparameters = learning rate 1e-4, momentum 0.9, 50k training steps, 10 NNs, 70 candidates per round
    Standard machine learning choices referenced from [19]; not fitted to the target Q value.
assumptions (4)
  • domain assumption The Q factor of a photonic crystal nanocavity is accurately computed by the 3D-FDTD method.
    All Q labels in the dataset are FDTD results; the record Q = 1.1e7 claim rests on FDTD accuracy (Section 3.1, Step II).
  • domain assumption High Q factors require mirror symmetries with respect to the central x and y axes, reducing the parameter space to 25 degrees of freedom.
    The optimization considers only symmetric air hole displacements (Section 3.1, Step I), so asymmetric structures are excluded from the search space.
  • ad hoc to paper A neural network with ReLU activations, dropout, and weight decay can learn a useful approximation to log10 Q(x) from a few thousand samples in a 25-dimensional space.
    This is the paper's core machine learning assumption, invoked in the learning phase (Section 3.2, Step IV) and necessary for the gradient-based candidate search.
  • ad hoc to paper The artificial loss terms (B) and (C) encourage exploration of unmeasured regions and prevent convergence to local maxima.
    This is the paper's design hypothesis for why strategies (A+B) and (A+C) outperform strategy (A), supported only by the single-run results in Section 3.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Iterative optimization of photonic crystal nanocavity designs by using deep neural networks." pith.science (2026). https://pith.science/paper/VU2PQI23

@misc{pith2026190803702,
  author       = {Pith},
  title        = {Pith review of: Iterative optimization of photonic crystal nanocavity designs by using deep neural networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VU2PQI23}},
  note         = {Machine review of arXiv:1908.03702}
}
read the original abstract

Devices based on two-dimensional photonic-crystal (2D-PC) nanocavities, which are defined by their air hole patterns, usually require a high quality (Q) factor to achieve high performance. We demonstrate that hole patterns with very high Q factors can be efficiently found by the iteration procedure consisting of: machine learning of the relation between the hole pattern and the corresponding Q factor, and new dataset generation based on the regression function obtained by machine learning. First a dataset comprising randomly generated cavity structures and their first principles Q factors is prepared. Then a deep neural network is trained using the initial dataset to obtain a regression function that approximately predicts the Q factors from the structural parameters. Several candidates for higher Q factors are chosen by searching the parameter space using the regression function. After adding these new structures and their first principles Q factors to the training dataset, the above process is repeated. As an example, a standard silicon-based L3 cavity is optimized by this method. A cavity design with a high Q factor exceeding 11 million is found within 101 iteration steps and a total of 8070 cavity structures. This theoretical Q factor is more than twice of the previously reported record values of the cavity designs detected by the evolutionary algorithm and the leaky mode visualization method. It is found that structures with higher Q factors can be detected within less iteration steps by exploring not only the parameter space near the present highest-Q structure but also that distant from the present dataset.

Figures

Figures reproduced from arXiv: 1908.03702 by the authors.

Figure 1
Figure 1. The iterative optimization proposed in this paper. Each cavity with the structural parameters selected in the preparation phase is represented by a unique high-dimensional vector . The accurate Q as a function of , , can be calculated by first principle approaches but is costly to compute. denotes the set of structures and consists of [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 4
Figure 4. Inter-structure distances between the structure with the highest Q factor in the present round (at 𝑏 ′ in the parameter space) and that in the previous rounds (at 𝑏 ) as a function of the size of the used training dataset. The inter-structure distance | 𝑏 ′ − 𝑏 | indicates how large the modifications in each round of optimization are. The results for the three different strategies (A), (A+B), and (A+C) are shown wit… view at source ↗
Figure 5
Figure 5. Air hole displacements of the optimized cavities with the highest Q factors found during the entire optimization for (a) strategy (A), (b) strategy (A+B), and (c) strategy (A+C). The circles represent the determined optimum positions of the air holes, and the arrows represent the displacement vectors with the scale shown by the red arrow. The electric field distributions of Ey are also plotted in color. The highest … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 30 canonical work pages

  1. [16]

    Auto ated opti izatio of photo ic crystal slab cavities,

    M. Minkov and V. Savo a, “Auto ated opti izatio of photo ic crystal slab cavities,” Sci. Rep. 4, 5124 (2015)

    work page 2015

  2. [17]

    I prove e t i the quality factors for photo ic crystal nanocavities via visualization of the leaky co po e ts,

    T. Naka ura, Y. Takahashi, Y. Ta aka, T. Asa o, a d S. Noda, “I prove e t i the quality factors for photo ic crystal nanocavities via visualization of the leaky co po e ts,” Opt. Express 24, 9541-9549 (2016)

    work page 2016

  3. [19]

    Opti izatio of photo ic crystal a ocavities based o deep lear i g,

    T. Asa o, S. Noda, “Opti izatio of photo ic crystal a ocavities based o deep lear i g,” Opt. Express. 26, 32704–32716 (2018)

    work page 2018

  4. [1]

    Trappi g a d e issio of photo s by a si gle defect i a photo ic ba dgap st ructure,

    S. Noda, A. Chuti a , a d M. I ada, “Trappi g a d e issio of photo s by a si gle defect i a photo ic ba dgap st ructure,” Nature 407, 608–610 (2000)

    work page 2000

  5. [2]

    High-Q photonic nanocavity in a two-di e sio al photo ic crystal,

    Y. Akahane, T. Asano, B.-S. So g, a d S. Noda, “High-Q photonic nanocavity in a two-di e sio al photo ic crystal,” Nature 425, 944-947 (2003)

    work page 2003

  6. [3]

    Ultra-high-Q photonic double-heterostructure nanocavity,

    B. S. So g, S. Noda, T. Asa o, a d Y. Akaha e, “Ultra-high-Q photonic double-heterostructure nanocavity,” Nat. Mater. 4, 207- 210 (2005)

    work page 2005

  7. [4]

    A alysis of the experi e tal Q factors (~ illio ) of photo ic crystal a ocavities,

    T. Asano, B.-S. So g, a d S. Noda, “A alysis of the experi e tal Q factors (~ illio ) of photo ic crystal a ocavities,” Opt. Express 14(5), 1996–2002 (2006)

    work page 2006

  8. [5]

    Ul trahigh- Q photonic crystal nanocavities realized by the local width odulatio of a li e defect,

    E. Kuramochi, M. Notomi, S. M itsugi, A. Shi ya, T. Ta abe, a d T. Wata abe, “Ul trahigh- Q photonic crystal nanocavities realized by the local width odulatio of a li e defect,” Appl. Phys. Lett. 88, 041112 (2006)

    work page 2006

Show all 31 references
  1. [6]

    Takahashi, H

    Y. Takahashi, H. Hagino, Y. Tanaka, B. S. Song, T. Asano, and S. Noda, “High-Q nanocavity with a 2- s photo lifeti e,“ Opt. Express 15, 17206-17213 (2007)

    2007

  2. [7]

    Ultrahigh -Q two -dimensional photonic crystal slab a ocavities i very thi barriers,

    E. Kura ochi, H. Ta iya a, T. Ta abe, A. Shi ya, a d M. Noto i, “Ultrahigh -Q two -dimensional photonic crystal slab a ocavities i very thi barriers,” Appl. Phys. Lett. 93, 111112 (2008)

    2008

  3. [8]

    Z. Han, X. Checoury, D. Néel, S. David, M. El Kurdi, a d P. Boucaud, “Opti ized desig for 2× 6 ultra-high Q silicon photonic crystal cavities,“ Opt. Co u . 283, 4387-4391 (2010)

    2010

  4. [9]

    Photonic crystal nanocavity with a Q-factor of ~9 million,

    H. Sekoguchi, Y. Takahashi, T. Asa o, a d S. Noda, “Photonic crystal nanocavity with a Q-factor of ~9 million,” Opt. Express 22, 916-924 (2014)

    2014

  5. [10]

    Photo ic crystal a ocavity with a Q factor exceedi g eleve illio ,

    T. Asa o, Y. Ochi, Y. Takahashi, K. Kishi oto, a d S. Noda, “Photo ic crystal a ocavity with a Q factor exceedi g eleve illio ,” Opt. Express 25, 1769-1777 (2017)

    2017

  6. [11]

    Photo ic Crystal Devices i Silico Photo ics,

    T. Asa o, S. Noda, “Photo ic Crystal Devices i Silico Photo ics,” Proc. IEEE 106, 1–13 (2018)

    2018

  7. [12]

    Mo e tu space design of high-Q photo ic crystal optical cavities,

    K. Sri ivasa a d O. Pai ter, “Mo e tu space design of high-Q photo ic crystal optical cavities,” Opt. Express 10, 670–684 (2002)

    2002

  8. [13]

    Ge eral recipe for desig i g photo ic crystal cavities,

    D. E glu d, I. Fush a , a d J. Vucković, “Ge eral recipe for desig i g photo ic crystal cavities,” Opt. Express 13, 5961–5975 (2005)

    2005

  9. [14]

    Design of Photonic Crystal Nanocavity With Q -Factor of ~ 10 9,

    Y. Ta aka, T. Asa o, a d S. Noda, “ Design of Photonic Crystal Nanocavity With Q -Factor of ~ 10 9,” J. Light. Tech ol. 26, 1532–1539 (2008)

    2008

  10. [15]

    Ge etically desig ed L3 photo ic crystal a ocavities with easured quality factor exceedi g o e illio ,

    Y. Lai, S. Pirotta, G. Urbi ati, D. Gerace, M. Mi kov, V. Savo a, A. Badolato, a d M. Galli, “Ge etically desig ed L3 photo ic crystal a ocavities with easured quality factor exceedi g o e illio ,” Appl. Phys. Lett. 104, 241101 (2014)

    2014

  11. [18]

    Photo ic crystal slab cavity si ulta eously opti ized for ultra-high Q / V and vertical radiatio coupli g,

    M. Mi kov, V. Savo a, a d D. Gerace, “Photo ic crystal slab cavity si ulta eously opti ized for ultra-high Q / V and vertical radiatio coupli g,” Appl. Phys. Lett. 111, 131104 (2017)

    2017

  12. [20]

    A alysis of high -Q photonic crystal L3 nanocavities designed by visualizatio of the leaky co po e ts,

    K. Mae o, Y. Takahashi, T. Naka ura, T. Asa o, S. Noda, “A alysis of high -Q photonic crystal L3 nanocavities designed by visualizatio of the leaky co po e ts,” Opt. Express 25, 367-376 (2017)

    2017

  13. [21]

    Ha dwritte Digit Recog itio with a Back-Propagatio Networks,

    Y. LeCu , B. Boser, J. S. De ker, D. He derso , R. E. Howard, W. Hubbard, a d L. D. Jacklel, “Ha dwritte Digit Recog itio with a Back-Propagatio Networks,” i Proceedi gs of Adva ces i Neural Information Processing Systems, pp. 396-404 (1990)

    1990

  14. [22]

    Deep Sparse Rectifier Neural Networks,

    X. Glorot, A. Bordes, a d Y. Be gio: “Deep Sparse Rectifier Neural Networks,” i Proceedi gs of Artificial I tellige ce a d Statistics, pp. 315-323 (2011)

    2011

  15. [23]

    Dropout: A Si ple Way to Preve t Neural Networks fro Overfitti g,

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, a d R. Salakhutdi ov, “Dropout: A Si ple Way to Preve t Neural Networks fro Overfitti g,” Jour al of Machi e Lear i g Research 15, 1929–1958 (2014)

    2014

  16. [24]

    A Si ple Weight Decay Ca I prove Ge eralizatio ,

    A. Krogh a d J. A. Hertz, “A Si ple Weight Decay Ca I prove Ge eralizatio ,” i Proceedi gs of Advances in Neural Information Processing Systems, pp. 950–957 (1991)

    1991

  17. [25]

    Lear i g represe tatio s by back -propagati g errors,

    D. E. Ru elhart, G. E. Hi to , a d R. J. Willia s, “Lear i g represe tatio s by back -propagati g errors,” Nature 323, 533– 536 (1986)

    1986

  18. [26]

    So e ethods of speedi g up the converge ce of iteratio ethods,

    B. T. Polyak, “So e ethods of speedi g up the converge ce of iteratio ethods,” USSR Co putatio al Mathe atics a d Mathematical Physics 4, 791-803 (1964)

    1964

  19. [27]

    Efficie t global opti izatio of expe sive black -box fu ctio s,

    D. R. Jones, M. Schonlau, and W. J. Welch. “Efficie t global opti izatio of expe sive black -box fu ctio s,” J. of Global Optimization 13, 455– 492 (1998)

    1998

  20. [28]

    Taki g the hu a out of the loop: A review of Bayesia opti izatio ,

    B. Shahriari, K. Swersky, Z. Wa g, R. P. Ada s, N. De Freitas, “Taki g the hu a out of the loop: A review of Bayesia opti izatio ,” Proc. IEEE 104, 148–175 (2016)

    2016

  21. [29]

    High Di e sio al Bayesia Opti izatio with Elastic Gaussia Process,

    S. Ra a, C. Li, S. Gupta, V. Nguye , S. Ve katesh, “High Di e sio al Bayesia Opti izatio with Elastic Gaussia Process,” Proc. of the 34th Int. Conf. on Mach. Learn. 70, 2883-2891 (2017)

    2017

  22. [30]

    Bayesia n optimization in a billion dimensions via random e beddi gs,

    Z. Wa g, F. Hutter, M. Zoghi, D. Matheso , N. De Freitas, “Bayesia n optimization in a billion dimensions via random e beddi gs,” J. Artif. Intell. Res. 55, 361–367 (2016)

    2016

  23. [31]

    Scalable Bayesia Opti izatio Usi g Deep Neural Networks,

    J. S oek et al., “Scalable Bayesia Opti izatio Usi g Deep Neural Networks,” Proc. of the 32 d I ter atio al Co fere ce o Machine Learning 37, (2015). (available at http://arxiv.org/abs/1502.05700)

    2015 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.