Pith. sign in

REVIEW 4 major objections 5 minor 45 references

Analysis of Photonic Circuit Losses with Machine Learning Techniques

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A simple linear regression model with variable selection predicts silicon nitride waveguide loss more accurately and interpretably than neural networks, random forests, and kernel methods, and points to wavelength as the dominant loss…

desk verdict Solid model benchmark with an overstated causal claim; the annealing validation is suggestive but under-reported. read the letter →

arxiv 2506.17999 v1 pith:JTW7Z3IH submitted 2025-06-22 physics.optics physics.app-phphysics.data-an

Adrian Nugraha Utama , Simon Chun Kiat Goh , Li Hongyu , Wang Xiangyu , Zhou Yanyan , Victor Leong , Manas Mukherjee This is my paper · ORCID
classification physics.opticsphysics.app-phphysics.data-an
keywords siliconnitridephotonicintegratedcircuitswaveguidelossmachinelearninglinearregressionvariableselectionannealingpropagation
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 claims that for predicting waveguide loss in a silicon nitride photonic platform, a simple linear regression model with variable selection is both more accurate and more interpretable than complex machine learning models such as random forests, support vector regression, and neural networks. The model's interpretability singled out wavelength as the dominant factor: at 1600 nm the propagation loss slope drops by 2.34 dB/cm relative to 1550 nm. Acting on that clue, the authors annealed devices at 1050°C for one hour and reduced the 1550 nm propagation loss by about 2 dB, matching the 1600 nm performance. The paper argues that the dataset's high linearity, rooted in the Beer-Lambert law, is why simple linear models win here.

What carries the argument

The load-bearing object is the linear regression with interaction terms, written as $y = b + \sum_i b_i x_i + \sum_{i\neq j} b_{ij} x_i x_j + \cdots + b_{12345} x_1 x_2 x_3 x_4 x_5$, where $y$ is the total waveguide insertion loss in dB and the features $x_i$ are length, width, CMP, wavelength, and polarization. Variable selection ranks coefficients by p-value and F-statistic per cross-validation fold, keeping the top $N=7$ terms; the tuned model contains the length main effect ($b_1 = 3.44$ dB/cm) and the length-by-wavelength interaction ($b_{14} = -2.34$ dB/cm), which together encode wavelength-dependent propagation loss. Nested cross-validation with 50 repetitions provides the benchmark that selects the upper-bound error as the comparison metric.

What would settle it

Measure propagation loss versus wavelength on a fresh set of devices spanning multiple wafers and process runs while recording wafer and run identifiers; if the about 2.3 dB/cm wavelength effect shrinks or vanishes once wafer or run effects are included in the model, the wavelength attribution that motivates the annealing experiment would be a confound rather than a physical cause.

Watch

Extended reading notes

Core claim

The central claim is that a linear regression model with interaction terms followed by p-value-based variable selection, fitted to 210 measured devices with five features (length, width, CMP, wavelength, polarization), achieves a lower and more consistent prediction error than any of the non-linear machine learning models tested, while exposing the physical drivers of loss. The key identified effect is the interaction term between waveguide length and wavelength at 1600 nm, with coefficient -2.34 dB/cm, meaning the propagation loss slope is substantially lower at 1600 nm than at 1550 nm. Attributing this to N-H absorption near 1520 nm, the authors demonstrate that post-fabrication annealing at 1050°C for one hour reduces propagation loss at 1550 nm by roughly 2 dB, at the cost of a slight increase in coupling loss and device non-uniformity.

Load-bearing premise

The analysis assumes that the five recorded features capture every systematic driver of loss, pooling 210 devices across wafers and process runs without tracking batch or run-to-run variation, so the wavelength coefficient could be entangled with unmeasured fabrication differences.

Editorial extensions

If this is right

  • On similar low-dimensional, physically linear datasets, a simple linear model with variable selection should be tried before deploying black-box models; it can match or beat them while remaining interpretable.
  • The wavelength effect at 1550 nm versus 1600 nm can be directly mitigated: post-fabrication annealing at 1050°C for 1 hour recovers about 2 dB/cm of propagation loss at C-band.
  • The model separates propagation-loss terms from coupling-loss terms via the length intercept, allowing designers to attribute loss contributions without extra measurements.
  • Benchmarking by the upper bound of the nested cross-validation RMSE range provides a conservative model-selection criterion for foundry process optimization.

Reading between the lines

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

  • The same variable-selection recipe could be applied to other foundry platforms or to richer feature sets; the interaction-term explosion noted in the paper suggests dimensionality reduction or regularization would be needed beyond about five features.
  • Because the annealing step slightly increased coupling loss and device non-uniformity, a follow-up could model annealing parameters (temperature, duration) as features to find a Pareto-optimal trade-off between propagation and coupling loss.
  • The pooling of 210 devices across wafers invites a check for wafer-level confounding; including wafer or run as a categorical feature would test whether wavelength or batch effects drive the 2.34 dB/cm interaction.
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

4 major / 5 minor

Summary. The manuscript applies seven supervised machine-learning models to 210 insertion-loss measurements of silicon nitride waveguides characterized by five design, fabrication, and operating parameters. Using 10-fold cross-validation and 50-repeat nested cross-validation, the authors report that linear-regression models, in particular a linear model with p-value-based variable selection, achieve low prediction error while remaining interpretable. The selected model's coefficient table identifies a length-wavelength interaction (b14 = -2.34 dB/cm for 1600 nm relative to 1550 nm) as the second-largest effect, which the authors interpret as evidence that operation at 1550 nm suffers higher propagation loss. They then perform post-fabrication annealing at 1050°C on an unspecified subset of devices and report a ~2 dB reduction in propagation loss at 1550 nm, attributed to removal of N-H absorption.

Significance. If the central claims held, the paper would be a useful demonstration that transparent, low-complexity models can outperform black-box regressors on well-controlled photonic fabrication data, and that ML-driven feature attribution can motivate a successful process improvement. The nested-cross-validation benchmarking is carefully executed and the use of standard open-source libraries makes the methods reproducible; the model-ranking comparisons are the strongest part. However, the causal attribution of the wavelength effect and the confirmatory annealing experiment currently lack the statistical controls needed to support the 'unique identification' claim. With additional experimental-design reporting and statistical analysis, the work could be a valuable case study for ML-guided photonics process development.

major comments (4)
  1. [Section 2.2 and Table 2] The dataset pools 210 devices, but the paper does not state the number of wafers or process runs or whether 1550 nm and 1600 nm are measured on the same physical waveguides. Since the key coefficient b14 = -2.34(4) dB/cm is a between-wavelength contrast in a pooled observational sample, it can absorb wafer-to-wafer or run-to-run variation in PECVD stoichiometry, roughness, or patterning. The statement that two identical devices come from different locations on the same wafer controls only within-wafer variability, not across-wafer or across-run variability. Please report the wafer/run distribution and pairing structure, and include wafer/run as a feature or random effect where possible, or the wavelength attribution used in Section 5 is not uniquely identified.
  2. [Section 5, Figure 6] The annealing experiment is presented as confirmation of the ML-derived wavelength mechanism, but the figure shows no error bars, the text does not specify how many or which devices were annealed, the 'pristine' and 'annealed' measurements are not shown to be paired on the same devices, there is no no-anneal control, and no statistical test is reported. A generic post-annealing reduction in propagation loss would produce a similar figure, so the ~2 dB improvement does not by itself validate the N-H absorption explanation. Please provide device counts, paired before/after data where feasible, a control group, and a statistical test such as a paired bootstrap or t-test.
  3. [Section 2.2] The exclusion of devices with >40 dB loss as 'defective' is implemented without a pre-specified criterion or sensitivity analysis. Because the final sample size and the wavelength/polishing balance depend on this threshold, please report the number and feature distribution of excluded devices, justify the threshold independently of the outcome, and show that the main coefficient estimates and the Section 5 conclusions are robust to plausible alternative thresholds. If the threshold preferentially removes 1550 nm devices, the wavelength comparison could be biased.
  4. [Section 4, Table 3] The claim that LR with variable selection offers 'a lower prediction error' is only weakly supported by the reported metrics: its NCV upper bound (1.33 dB) is slightly worse than LR with interaction (1.31 dB), and its range substantially overlaps LASSO and SVR. The argument that overlapping confidence intervals do not prove equivalence is not a substitute for a paired comparison. Please report paired differences in per-repetition NCV RMSE across models, with a suitable significance test, or state explicitly that the top models are statistically indistinguishable in prediction error.
minor comments (5)
  1. [Figure 6] The label 'Annnealed' is a typo and should read 'Annealed'.
  2. [Section 6] In the Beer-Lambert paragraph, the waveguide length is called x0 while Table 1 defines it as x1, and the phrase 'x1 to x4' should be 'x2 to x5'.
  3. [Section 2.2] The statement that the data are 'almost equally distributed across the different features' would benefit from exact counts, especially the number of devices per wavelength and per polishing group.
  4. [Section 3.3] In Figure 2, the x-axis label 'Parameter' should explicitly identify the LASSO regularization parameter alpha, and the axis scale should be marked as logarithmic.
  5. [Section 3.6] The ANN hyperparameter is written as 'n_layer' in the text but 'n_layers' in the architecture description; please use consistent notation.

Circularity Check

0 steps flagged · score 2.0 of 10

No material circularity: model benchmarking uses held-out nested cross-validation, and the wavelength claim is tested by an independent annealing experiment; only a minor non-load-bearing self-citation appears.

full rationale

The paper's central derivation chain is not circular. The regression models are explicitly fit on training folds and evaluated on held-out validation folds via 10-fold nested cross-validation with 50 repetitions (Section 4), so the reported RMSE comparisons are genuine out-of-sample benchmarks rather than refits of the training data. The wavelength attribution is read from the refitted coefficient b14 in Table 2, but Section 5 then tests that attribution with a new post-fabrication annealing experiment performed at 1550 nm on devices not used to derive the model; the observed ~2 dB propagation-loss reduction is independent evidence for the N-H absorption mechanism. No parameter is fit to the annealed outcome and then called a prediction. The only self-citation is Ref. [21], used to justify the waveguide cutback technique; cutback is a standard external method and this citation does not carry the paper's central benchmarking or mitigation claims. Potential weaknesses such as pooling devices across wafers/runs and the absence of error bars in Figure 6 are threats to causal identification and external validity, not cases where a result reduces by definition to its inputs. Accordingly, the paper is self-contained with respect to its main derivation chain and receives a low score reflecting only the minor self-citation.

Assumptions & free parameters 15 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a fitted linear model with seven selected coefficients, plus a collection of tuned hyperparameters for each ML method. The physical interpretation relies on the known N-H absorption mechanism from the cited literature, and the data analysis assumes that unmeasured batch effects are negligible.

free parameters (15)
  • Intercept b = 4.3(1) dB
    Fitted constant in the tuned LR with variable selection (Table 2).
  • WG Length coefficient b1 = 3.44(3) dB/cm
    Fitted propagation loss slope in the tuned model (Table 2).
  • Length x Wavelength [1600 nm] coefficient b14 = -2.34(4) dB/cm
    Fitted interaction term that drives the wavelength insight (Table 2).
  • Length x Polarization [TM] coefficient b15 = -1.05(2) dB/cm
    Fitted interaction term (Table 2).
  • Length x Wavelength [1600 nm] x Polarization [TM] coefficient b145 = 0.85(2) dB/cm
    Fitted three-way interaction term (Table 2).
  • Length x Polishing [with CMP] coefficient b13 = -0.30(1) dB/cm
    Fitted interaction term (Table 2).
  • WG Width [1.5 um] coefficient b2 = 0.4(1) dB
    Fitted coefficient for waveguide width (Table 2).
  • Number of selected variables N = 7
    Tuned by minimizing cross-validated RMSE (Figure 1).
  • LASSO regularization alpha = 0.01
    Tuned regularization strength (Figure 2).
  • Maximum tree depth for decision tree and random forest = 8
    Tuned hyperparameter (Figure 3).
  • SVR epsilon = 1.0
    Tuned error-free region width (Figure 4).
  • SVR C = 30
    Tuned regularization parameter (Figure 4).
  • KRR regularization parameter (1/(2 alpha)) = 30
    Tuned equivalent of C for kernel ridge regression (Figure 4).
  • ANN number of layers = 3
    Tuned network architecture (Figure 5).
  • ANN number of neurons = 64
    Tuned network width (Figure 5).
assumptions (6)
  • standard math Least squares fitting and p-value/F-statistic rankings are valid proxies for variable importance.
    Invoked in Sections 3.1 and 3.2 for model fitting and variable selection.
  • domain assumption Insertion loss decomposes linearly into propagation loss and coupling loss with waveguide length via the cutback technique.
    Stated in Section 2.1 and used to interpret the length coefficient as propagation loss.
  • domain assumption Waveguide loss depends linearly on length with no self-interacting terms, following the Beer-Lambert law.
    Used in Section 3.1 and Discussion to justify excluding quadratic terms.
  • domain assumption N-H stretching bond absorption peaks near 1520 nm and annealing above 1000°C reduces this absorption.
    Taken from refs [38,39,40,41] in Section 5 to motivate the annealing experiment.
  • ad hoc to paper Devices with >40 dB loss are defective and can be excluded from analysis.
    Introduced in Section 2.2 without a pre-specified statistical rule, affecting less than 5% of devices.
  • domain assumption The dataset is representative and features are almost equally distributed, with duplicate devices capturing fabrication variability.
    Stated in Section 2.2; no batch or wafer-level random effects are modeled.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Analysis of Photonic Circuit Losses with Machine Learning Techniques." pith.science (2026). https://pith.science/paper/JTW7Z3IH

@misc{pith2026250617999,
  author       = {Pith},
  title        = {Pith review of: Analysis of Photonic Circuit Losses with Machine Learning Techniques},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JTW7Z3IH}},
  note         = {Machine review of arXiv:2506.17999}
}
read the original abstract

Low-loss waveguides enable efficient light delivery in photonic circuits, which are essential for high-speed optical communications and scalable implementations of photonic quantum technologies. We study the effects of several fabrication and experimental parameters on the waveguide losses of a silicon nitride integrated photonics platform using various machine learning techniques. Compared to more complex machine learning algorithms, our results show that a simple linear regression model with variable selection offers a lower prediction error with high interpretability. The high interpretability, along with our domain knowledge, led to unique identification of fabrication process dependencies on the final outcome. With these insights, we show that by improving the process flow, it is possible to improve the loss by mitigating the cause in a real experiment.

Figures

Figures reproduced from arXiv: 2506.17999 by the authors.

Figure 1
Figure 1. The performance of the LR with variable selection model for different numbers [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. The training and validation RMSE (top) with respect to the regularization [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. The training and validation RMSE of the decision tree and random forest [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The training and validation RMSE of the SVR and KRR models with respect [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: The training (left) and validation (right) RMSE heat maps of our ANN model, [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The measured propagation loss (left) and coupling loss (right) of several pristine [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 43 canonical work pages

  1. [1]

    Review of silicon photonics technology and platform development,

    S. Y. Siew, B. Li, F. Gao,et al., “Review of silicon photonics technology and platform development,” J. Light. Technol.39, 4374–4389 (2021)

    work page 2021

  2. [2]

    Wafer-scale demonstration of low-loss (∼0.43 dB/cm), high-bandwidth (>38 GHz), silicon photonics platform operating at the C-band,

    J. X. B. Sia, X. Li, J. Wang,et al., “Wafer-scale demonstration of low-loss (∼0.43 dB/cm), high-bandwidth (>38 GHz), silicon photonics platform operating at the C-band,” IEEE Photonics J.14, 1–9 (2022)

    work page 2022

  3. [3]

    High-yield, wafer-scale fabrication of ultralow-loss, dispersion-engineered silicon nitride photonic circuits,

    J. Liu, G. Huang, R. N. Wang,et al., “High-yield, wafer-scale fabrication of ultralow-loss, dispersion-engineered silicon nitride photonic circuits,” Nat. communications12, 2236 (2021)

    work page 2021

  4. [4]

    Foundry manufacturing of tight-confinement, dispersion-engineered, ultralow-loss silicon nitride photonic integrated circuits,

    Z. Ye, H. Jia, Z. Huang,et al., “Foundry manufacturing of tight-confinement, dispersion-engineered, ultralow-loss silicon nitride photonic integrated circuits,” Photonics Res.11, 558–568 (2023)

    work page 2023

  5. [5]

    Silicon quantum photonics,

    J. W. Silverstone, D. Bonneau, J. L. O’Brien, and M. G. Thompson, “Silicon quantum photonics,” IEEE J. Sel. Top. Quantum Electron.22, 390–402 (2016)

    work page 2016

  6. [6]

    Silicon nitride passive and active photonic integrated circuits: trends and prospects,

    C. Xiang, W. Jin, and J. E. Bowers, “Silicon nitride passive and active photonic integrated circuits: trends and prospects,” Photonics Res.10, A82–A96 (2022)

    work page 2022

  7. [7]

    Capturing the effects of spatial process variations in silicon photonic circuits,

    Y. Xing, J. Dong, U. Khan, and W. Bogaerts, “Capturing the effects of spatial process variations in silicon photonic circuits,” ACS Photonics10, 928–944 (2022)

    work page 2022

  8. [8]

    Process variation in silicon photonic devices,

    X. Chen, M. Mohamed, Z. Li,et al., “Process variation in silicon photonic devices,” Appl. optics52, 7638–7647 (2013)

    work page 2013

Show all 45 references
  1. [9]

    Optical quantum computing,

    J. L. O’brien, “Optical quantum computing,” Science318, 1567–1570 (2007)

    2007

  2. [10]

    Nanoscale low crosstalk photonic crystal integrated sensor array,

    D. Yang, H. Tian, and Y. Ji, “Nanoscale low crosstalk photonic crystal integrated sensor array,” IEEE Photonics J.6, 1–7 (2014)

    2014

  3. [11]

    Performancepredictionforsiliconphotonicsintegratedcircuitswithlayout-dependent correlated manufacturing variability,

    Z.Lu,J.Jhoja,J.Klein, et al.,“Performancepredictionforsiliconphotonicsintegratedcircuitswithlayout-dependent correlated manufacturing variability,” Opt. express25, 9712–9733 (2017)

    2017

  4. [12]

    Modelling fabrication variability in silicon photonic devices

    M. A. Hamdani and G. Qazi, “Modelling fabrication variability in silicon photonic devices.” Photonic Materials: Recent Adv. Emerg. Appl. p. 265 (2023)

    2023

  5. [13]

    Advances in machine learning for large-scale manufacturing of photonic circuits,

    K. Yadav, S. Bidnyk, and A. Balakrishnan, “Advances in machine learning for large-scale manufacturing of photonic circuits,” inEPJ Web of Conferences,vol. 287 (EDP Sciences, 2023), p. 13004

    2023

  6. [14]

    Machine learning for mask/wafer hotspot detection and mask synthesis,

    Y. Lin, X. Xu, J. Ou, and D. Z. Pan, “Machine learning for mask/wafer hotspot detection and mask synthesis,” in Photomask Technology 2017,vol. 10451 (SPIE, 2017), pp. 72–84

    2017

  7. [15]

    Use of neural network to model the deposition rate of PECVD-silicon nitride films,

    B. Kim, K. Park, and D. Lee, “Use of neural network to model the deposition rate of PECVD-silicon nitride films,” Plasma Sources Sci. Technol.14, 83 (2005)

    2005

  8. [16]

    Regression methods for prediction of PECVD silicon nitride layer thickness,

    H. Purwins, A. Nagi, B. Barak,et al., “Regression methods for prediction of PECVD silicon nitride layer thickness,” in 2011 IEEE International Conference on Automation Science and Engineering, (IEEE, 2011), pp. 387–392

    2011

  9. [17]

    Neural network modeling of inter-characteristics of silicon nitride film deposited by using a plasma-enhanced chemical vapor deposition,

    S. J. Lee, B. Kim, and S. W. Baik, “Neural network modeling of inter-characteristics of silicon nitride film deposited by using a plasma-enhanced chemical vapor deposition,” Expert Syst. with Appl.38, 11437–11441 (2011)

    2011

  10. [18]

    Evaluating regression-based techniques for modelling fabrication variations in silicon photonic waveguides,

    A. E. James, A. Wang, S. Wang, and K. Bergman, “Evaluating regression-based techniques for modelling fabrication variations in silicon photonic waveguides,” inApplications of Machine Learning 2021, vol. 11843 (SPIE, 2021), pp. 18–23

    2021

  11. [19]

    Random forest modelling as a tool for propagation and bend excess loss minimization on silicon nitride waveguide platforms,

    J. W. Hinum-Wagner, S. M. Hoermann, G. Feigl,et al., “Random forest modelling as a tool for propagation and bend excess loss minimization on silicon nitride waveguide platforms,” inMachine Learning in Photonics, vol. 13017 (SPIE, 2024), pp. 216–230

    2024

  12. [20]

    Deep learning models for effective refractive indices in silicon nitride waveguides,

    G. Alagappan and C. E. Png, “Deep learning models for effective refractive indices in silicon nitride waveguides,” J. Opt.21, 035801 (2019)

    2019

  13. [21]

    Improved waveguide surface roughness by foundry-processing techniques for enhanced light delivery to integrated ion trap for quantum computing platforms,

    S. C. Goh, A. N. Utama, H. Yu,et al., “Improved waveguide surface roughness by foundry-processing techniques for enhanced light delivery to integrated ion trap for quantum computing platforms,” inOptical Interconnects XXIV,vol. 12892 (SPIE, 2024), pp. 87–93

    2024

  14. [22]

    A study of cross-validation and bootstrap for accuracy estimation and model selection,

    R. Kohaviet al., “A study of cross-validation and bootstrap for accuracy estimation and model selection,” inIjcai, vol. 14(2) (Montreal, Canada, 1995), pp. 1137–1145

    1995

  15. [23]

    Linearregression,

    G.James,D.Witten,T.Hastie, et al.,“Linearregression,”in An introduction to statistical learning: With applications in python, (Springer, 2023), pp. 69–134

    2023

  16. [24]

    A survey of feature selection and feature extraction techniques in machine learning,

    S. Khalid, T. Khalil, and S. Nasreen, “A survey of feature selection and feature extraction techniques in machine learning,” in2014 science and information conference, (IEEE, 2014), pp. 372–378

    2014

  17. [25]

    Regression modeling strategies,

    F. E. Harrell, “Regression modeling strategies,” R package version pp. 6–2 (2012)

    2012

  18. [26]

    Generalized linear models,

    J. A. Nelder and R. W. Wedderburn, “Generalized linear models,” J. Royal Stat. Soc. Ser. A: Stat. Soc.135, 370–384 (1972)

    1972

  19. [27]

    Regression shrinkage and selection via the lasso,

    R. Tibshirani, “Regression shrinkage and selection via the lasso,” J. Royal Stat. Soc. Ser. B: Stat. Methodol.58, 267–288 (1996)

    1996

  20. [28]

    Induction of decision trees,

    J. R. Quinlan, “Induction of decision trees,” Mach. learning1, 81–106 (1986)

    1986

  21. [29]

    Random forests,

    L. Breiman, “Random forests,” Mach. learning45, 5–32 (2001)

    2001

  22. [30]

    Scikit-learn: Machine learning in python,

    F. Pedregosa, G. Varoquaux, A. Gramfort,et al., “Scikit-learn: Machine learning in python,” J. machine learning research12, 2825–2830 (2011)

    2011

  23. [31]

    A tutorial on support vector regression,

    A. J. Smola and B. Schölkopf, “A tutorial on support vector regression,” Stat. computing14, 199–222 (2004)

    2004

  24. [32]

    Kernel ridge regression,

    V. Vovk, “Kernel ridge regression,” inEmpirical inference: Festschrift in honor of vladimir n. vapnik, (Springer, 2013), pp. 105–116

    2013

  25. [33]

    TensorFlow: Large-scale machine learning on heterogeneous systems,

    M. Abadi, A. Agarwal, P. Barham,et al., “TensorFlow: Large-scale machine learning on heterogeneous systems,” (2015). Software available from tensorflow.org

    2015

  26. [34]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 (2014)

    2014 arXiv

  27. [35]

    No unbiased estimator of the variance of k-fold cross-validation,

    Y. Bengio and Y. Grandvalet, “No unbiased estimator of the variance of k-fold cross-validation,” Adv. Neural Inf. Process. Syst.16(2003)

    2003

  28. [36]

    On estimating model accuracy with repeated cross-validation,

    G. Vanwinckelen and H. Blockeel, “On estimating model accuracy with repeated cross-validation,” inBeneLearn 2012: Proceedings of the 21st Belgian-Dutch conference on machine learning, (2012), pp. 39–44

    2012

  29. [37]

    Cross-validation: what does it estimate and how well does it do it?

    S. Bates, T. Hastie, and R. Tibshirani, “Cross-validation: what does it estimate and how well does it do it?” J. Am. Stat. Assoc.119, 1434–1445 (2024)

    2024

  30. [38]

    Plasma enhanced chemical vapor deposition silicon oxynitride optimized for application in integrated optics,

    K. Wörhoff, A. Driessen, P. Lambeck,et al., “Plasma enhanced chemical vapor deposition silicon oxynitride optimized for application in integrated optics,” Sensors Actuators A: Phys.74, 9–12 (1999)

    1999

  31. [39]

    Comparative investigation of hydrogen bonding in silicon based PECVD grown dielectrics for optical waveguides,

    F. Ay and A. Aydinli, “Comparative investigation of hydrogen bonding in silicon based PECVD grown dielectrics for optical waveguides,” Opt. materials26, 33–46 (2004)

    2004

  32. [40]

    Low loss Si3N4-SiO2 optical waveguides on Si,

    C. H. Henry, R. F. Kazarinov, H. J. Lee,et al., “Low loss Si3N4-SiO2 optical waveguides on Si,” Appl. optics26, 2621–2624 (1987)

    1987

  33. [41]

    Nonlinear silicon nitride waveguides based on a PECVD deposition platform,

    L. Wang, W. Xie, D. Van Thourhout,et al., “Nonlinear silicon nitride waveguides based on a PECVD deposition platform,” Opt. express26, 9645–9654 (2018)

    2018

  34. [42]

    CMOS compatible monolithic multi-layer Si3N4-on-SOI platform for low-loss high performance silicon photonics dense integration,

    Y. Huang, J. Song, X. Luo,et al., “CMOS compatible monolithic multi-layer Si3N4-on-SOI platform for low-loss high performance silicon photonics dense integration,” Opt. Express22, 21859–21865 (2014)

    2014

  35. [43]

    Material and optical properties of low-temperature NH3-free PECVD SiNx layers for photonic applications,

    T. D. Bucio, A. Z. Khokhar, C. Lacava,et al., “Material and optical properties of low-temperature NH3-free PECVD SiNx layers for photonic applications,” J. Phys. D: Appl. Phys.50, 025106 (2016)

    2016

  36. [44]

    Asurveyofdeepneuralnetworkarchitecturesandtheirapplications,

    W.Liu,Z.Wang,X.Liu, et al.,“Asurveyofdeepneuralnetworkarchitecturesandtheirapplications,”Neurocomputing 234, 11–26 (2017)

    2017

  37. [45]

    The Beer-Lambert law,

    D. F. Swinehart, “The Beer-Lambert law,” J. chemical education39, 333 (1962)

    1962

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.