Pith. sign in

REVIEW 4 major objections 4 minor 39 references

Hybrid Quantum-Classical Learning of Nonlinear Entanglement Witnesses via Continuous-Variable Quantum Neural Networks

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A hybrid continuous-variable quantum neural network can be trained to act as a nonlinear entanglement witness, classifying separable versus entangled states with above 99% accuracy and keeping that accuracy in three-mode problems where clas

desk verdict The two-mode demo and the theory are fine; the three-mode gap is the whole story and it rests on a baseline comparison the paper contradicts itself about and never actually reports. read the letter →

arxiv 2509.05924 v1 pith:FCE36GWH submitted 2025-09-07 quant-ph

Mohammad Rezaei Shokou , Hossein Davoodi Yeganeh This is my paper · ORCID
classification quant-ph MSC 81P4081P68 PACS 03.65.Ud03.67.Mn
keywords entanglementwitnesscontinuous-variablequantumneuralnetworksmachinelearninghybridquantum-classicalmultipartitephotonlossuniversalapproximationstateclassification
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 tries to establish that a continuous-variable quantum neural network (CV-QNN) can be trained as a nonlinear entanglement witness that classifies quantum states as separable or entangled directly from measurement data. The authors report above 99% classification accuracy on two- and three-mode datasets covering Gaussian and non-Gaussian, pure and mixed states, with a performance gap that widens as the system grows: classical SVM and MLP baselines drop to roughly 75% in the three-mode case. The paper also proves that, when the measurement readout is informationally complete, the hybrid architecture can approximate any continuous witness-like functional on compact sets of states. A sympathetic reader would care because separability is NP-hard in general, so a trainable, hardware-native witness that scales better than classical models could offer a practical route to entanglement certification on near-term photonic devices.

What carries the argument

The core object is the learned witness functional W(ρ)=f_Φ(g_Θ(ρ)). The quantum stage is a two-layer CV-QNN built from four gate blocks per layer: an interferometer that mixes modes, single-mode squeezing and displacement that act as affine scaling and bias, and a Kerr gate that supplies non-Gaussian nonlinearity. The readout is an informationally complete POVM, implemented in simulation by measuring Fock-basis probabilities after several fixed pre-measurement unitaries; injectivity of this feature map is what lets the classical head distinguish any two distinct states. The classical head is a small MLP that maps the concatenated probability vector to a scalar witness value. The universal-ap

What would settle it

Train the SVM and MLP on the same shot-noise-limited features from the untrained IC measurement circuit used for the quantum head; if either reaches about 99% accuracy on three-mode states, the claimed scalable performance gap disappears.

Watch

Extended reading notes

Core claim

The central claim is that a variational continuous-variable circuit composed of interferometers, squeezers, displacements, and Kerr gates, followed by an informationally complete measurement and a small classical neural head, learns a nonlinear witness functional W(ρ)=f_Φ(g_Θ(ρ)) that separates entangled from separable states with high accuracy. In numerical experiments, the model achieves 99.02% accuracy on two-mode states and 99.00% on three-mode states, while the best classical baseline drops from about 98% to 76%. The paper further claims that when the measurement map is injective, the witness functional class is dense in the continuous functions on any compact set of states, so the arch

Load-bearing premise

The classical baselines must receive genuinely equivalent information and noise conditions; the paper reports the full-density-matrix engineered-feature results but not the promised matched-feature baseline, so the fairness of the comparison is the load-bearing assumption.

Editorial extensions

If this is right

  • A learned nonlinear witness can classify both Gaussian and non-Gaussian, pure and mixed two- and three-mode states with over 99% accuracy using a finite 1000-shot readout.
  • Scaling from two modes to three modes leaves the CV-QNN accuracy essentially unchanged (99.0%) while the best classical baseline drops to roughly 76%, giving an empirical scalable gap.
  • At 10% per-layer photon loss, the three-mode quantum model still classifies above 97%, outperforming noiseless classical baselines across the tested loss range.
  • With an informationally complete readout, the hybrid architecture is dense in the continuous functions on any compact set of states, so it can in principle represent any continuous witness functional.

Reading between the lines

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

  • If the classical baselines were instead given the same shot-noise-limited features from an untrained circuit (the matched-feature scenario the paper describes but does not report), the performance gap would likely shrink; the claimed quantum-native advantage may reduce to an information-asymmetry effect.
  • The universal-approximation theorem covers expressivity, not trainability; nothing in the paper rules out barren-plateau-like behavior or poor sample complexity as the number of modes grows.
  • The same pipeline could be pointed at harder certification tasks, such as bound entanglement or non-Gaussian state verification, where no simple analytical witness exists; the paper does not test these.
  • A direct hardware test on a photonic chip with loss could settle whether the loss robustness persists when the noise is not simulated by the same model used for training.
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 / 4 minor

Summary. The manuscript proposes a hybrid quantum-classical model for learning nonlinear entanglement witnesses from continuous-variable quantum data. The architecture combines a CV-QNN (interferometer, squeezers, displacements, Kerr gates) with an informationally complete POVM readout and a classical MLP head. The authors report numerical experiments on two- and three-mode state datasets (Gaussian, non-Gaussian, pure, mixed), claiming >99% accuracy with a 'robust scalable performance gap' over SVM and MLP baselines, robustness to photon loss under finite shots, and a universal approximation theorem for continuous witness functionals when the measurement is informationally complete.

Significance. If the empirical claims were well supported, the work would provide a useful benchmark for quantum machine learning in entanglement certification. The paper contains reproducible architectural details, finite-shot simulation, and statistical bootstrap CIs. However, the central scalability claim rests on a baseline comparison that is internally inconsistent and partially missing; the photon-loss comparison is asymmetric; and the theoretical theorem is a corollary of the classical universal approximation theorem. As presented, the significance of the claimed advantage cannot be assessed.

major comments (4)
  1. [III.C.2, IV.B, Conclusion] The paper promises a 'Direct Comparison (Matched Features)' in §III.C.2 but never reports it. The Results (Table II, §IV.B) say classical baselines were given 'a comprehensive set of engineered features' (purity, entropies, all bipartite negativities, flattened density matrix), while the Conclusion says they had 'the same post-measurement information as the quantum model's classical head.' These are different inputs. The matched-feature experiment is essential to isolate the effect of training the quantum circuit; its absence makes the reported performance gap impossible to verify.
  2. [Algorithm 1, §III.C.2] The entangled labels in Algorithm 1 are defined by requiring all bipartite negativities to be positive (step 9). The engineered feature vector in §III.C.2 includes those same negativities. A simple threshold on the maximum negativity would then classify the training/test sets almost perfectly, contradicting the reported 76% (SVM) and 74.5% (MLP) three-mode accuracies. Either the engineered features were not actually supplied to the baselines, or the dataset labels are not consistent with the feature description. Neither case supports the claimed comparison.
  3. [IV.C, Fig. 6] The photon-loss experiment compares the noisy CV-QNN (with per-layer loss and Nshots=1000 shot noise) against noiseless classical baselines (SVM/MLP). Classical baselines are not subjected to the same photon-loss channel or shot noise. The observed gap under noise may therefore be an artifact of asymmetric noise treatment rather than a native advantage. A matched-noise comparison is required.
  4. [Theorem II.1, Appendix A] The theorem is essentially the classical universal approximation theorem applied to the image of an IC-POVM. The proof fixes a single Θ* with injective gΘ*, then invokes UAT for MLPs; the variational circuit U(Θ) and its trainability play no role. This does not establish any approximation advantage specific to CV-QNNs. The statement should be substantially qualified or replaced with a result that uses the variational degrees of freedom.
minor comments (4)
  1. [IV.B] The statement that non-overlapping 95% CIs imply p<0.001 is not justified; bootstrap CI non-overlap does not directly give a p-value. Please either report a proper significance test or remove the p-value.
  2. [Appendix C] The reference to 'Algorithm 7' should be 'Algorithm 1'.
  3. [II.C, Eq. (7)] The symbol K is used both for the number of IC measurement settings and for the Kerr gate; consider different notation.
  4. [Throughout] The paper would benefit from stating explicitly whether the classical baselines were given access to the full density matrix in the three-mode experiments; the text is inconsistent on this point.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central empirical benchmark is independent and the universal-approximation theorem, while low in novelty, is not circular.

full rationale

The paper's central claim is an empirical supervised-learning benchmark: models are trained on one split and evaluated on a held-out test split (Algorithm 1, Section III.D, Tables I-II). There is no fitted-parameter-renamed-as-prediction and no quantity is defined in terms of the target result. The universal-approximation theorem (Theorem II.1 and Appendix A) is not circular: injectivity of an IC-POVM and the classical universal approximation theorem are independent premises that entail density in C(X); the proof is explicit and does not assume its conclusion. The result is admittedly low in novelty because the variational unitary plays no real role in the proof, but low novelty is not circular derivation. The CV-QNN architecture is attributed to Killoran et al. (Ref. [18]), an external and code-available prior work, not to the present authors, so there is no load-bearing self-citation chain. The unperformed matched-feature classical baseline promised in Section III.C.2 and the asymmetric noisy-quantum vs. noiseless-classical photon-loss comparison are reproducibility and fairness concerns, not circular reductions: they do not make any equation equal its own input. The paper also candidly notes in Section IV.D.3 that a formal quantum-advantage proof remains open, which is consistent with a non-circular empirical claim. Overall, no circular step can be exhibited from the text.

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

The central claims rest on trained circuit parameters, truncated Fock spaces, an unspecified IC readout, and a fragile baseline-comparison assumption. The theorem depends on the informationally complete assumption, which is stated but not instantiated. No new physical entities are introduced.

free parameters (5)
  • CV-QNN gate parameters (rotations, beamsplitters, squeezers, displacements, Kerr) = 24 (2-mode), 42 (3-mode), L=2 layers
    Initialized as N(0,0.1) and optimized by Adam; these are the trained weights of the quantum circuit. The performance claims depend on them being fitted to the datasets.
  • Classical MLP head weights = two hidden layers of 64 units (also 128/64 in 3-mode classical MLP baseline)
    Optimized by backpropagation; standard trainable parameters of the final classifier.
  • Fock cutoff d and truncation = d=4 (2-mode), d=3 (3-mode)
    Chosen for simulation tractability; the central results are computed inside this truncated Hilbert space and depend on the accompanying trace-penalty regularizer.
  • Trace-penalty coefficient gamma = not reported
    Appears in loss function (9) but no value is given; a free tuning constant that the results depend on.
  • Number of IC measurement settings K and pre-measurement unitaries {V_k} = not reported
    The informationally complete readout is central to both the method and the theorem, but K and the specific unitaries are not specified in the paper.
assumptions (5)
  • domain assumption Negativity over all bipartite splits is a correct labeling rule for the generated entangled states
    Algorithm 1 verifies entanglement by computing bipartite negativities and discards states with all non-positive negativity. For three-mode states this omits some entangled classes, but the dataset is restricted to states where negativity is positive.
  • domain assumption The fixed pre-measurement unitaries {V_k} plus Fock-basis measurement form an informationally complete POVM on the truncated Hilbert space
    Theorem II.1 requires an injective feature map, but the paper asserts the IC property without specifying K or {V_k}, so the core theoretical premise is not instantiated.
  • domain assumption The classical baselines are strong and fairly configured
    The central performance-gap claim depends on the baselines being given fair and powerful features. The paper's internal inconsistency about engineered vs matched features makes this assumption fragile.
  • domain assumption Fock-space truncation with a trace-penalty regularizer faithfully approximates the infinite-dimensional CV dynamics
    All simulations use finite cutoff d with a trace penalty; convergence in d is not studied, yet the results are presented as relevant to real photonic systems.
  • domain assumption Strawberry Fields Fock backend correctly simulates the gates
    All numerical claims rely on this tool; no independent implementation or cross-check is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hybrid Quantum-Classical Learning of Nonlinear Entanglement Witnesses via Continuous-Variable Quantum Neural Networks." pith.science (2026). https://pith.science/paper/FCE36GWH

@misc{pith2026250905924,
  author       = {Pith},
  title        = {Pith review of: Hybrid Quantum-Classical Learning of Nonlinear Entanglement Witnesses via Continuous-Variable Quantum Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FCE36GWH}},
  note         = {Machine review of arXiv:2509.05924}
}
read the original abstract

A major challenge in quantum information is characterizing entanglement, for which entanglement witnesses offer effective means of detecting quantum correlations. We introduce a hybrid quantum-classical framework that learns a nonlinear entanglement witness directly from quantum data using continuous-variable quantum neural networks (CV-QNNs). Our architecture combines variational interferometers, squeezers and non-Gaussian Kerr gates with a small classical neural head to output a scalar witness value. Numerical simulations were conducted on two- and three-mode families, including Gaussian and non-Gaussian states in both pure and mixed forms. We observed over 99% classification accuracy and a robust performance gap compared to strong classical baselines, especially when scaling from two to three modes. Robustness to photon loss is further quantified under a finite number of measurement shots. On the theory side, we show that when the quantum measurement stage is informationally complete, the hybrid model can approximate any continuous witness-like functional on compact sets of states.Our findings highlight CV-QNNs as a promising framework for data-driven quantum state characterization and propose specific benchmarks where near-term photonic platforms offer tangible advantages.

Figures

Figures reproduced from arXiv: 2509.05924 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 29 canonical work pages

  1. [1]

    Quantum entanglement.Reviews of modern physics, 81(2):865–942, 2009

    Ryszard Horodecki, Pawe/suppress l Horodecki, Micha/suppress l Horodecki, and Karol Horodecki. Quantum entanglement.Reviews of modern physics, 81(2):865–942, 2009

    work page 2009

  2. [2]

    Cambridge university press, 2010

    Michael A Nielsen and Isaac L Chuang.Quantum computation and quantum information. Cambridge university press, 2010

    2010

  3. [3]

    Classical complexity and quantum entanglement.Journal of Computer and System Sciences, 69(3):448–484, 2004

    Leonid Gurvits. Classical complexity and quantum entanglement.Journal of Computer and System Sciences, 69(3):448–484, 2004

    work page 2004

  4. [4]

    Separability criterion for density matrices.Physical Review Letters, 77(8):1413, 1996

    Asher Peres. Separability criterion for density matrices.Physical Review Letters, 77(8):1413, 1996

    work page 1996

  5. [5]

    Bell inequalities and the separability criterion.Physics Letters A, 271(5-6):319–326, 2000

    Barbara M Terhal. Bell inequalities and the separability criterion.Physics Letters A, 271(5-6):319–326, 2000

    work page 2000

  6. [6]

    Nonlinear power-like and svd-like iterative schemes with applications to entangled bipartite rank-1 approximation.SIAM Journal on Scientific Computing, 43(5):S448–S474, 2021

    Moody T Chu and Matthew M Lin. Nonlinear power-like and svd-like iterative schemes with applications to entangled bipartite rank-1 approximation.SIAM Journal on Scientific Computing, 43(5):S448–S474, 2021

    work page 2021

  7. [7]

    Supervised learning with quantum computers.Quantum science and technology, 17, 2018

    Maria Schuld and Francesco Petruccione. Supervised learning with quantum computers.Quantum science and technology, 17, 2018

    work page 2018

  8. [8]

    Springer, 2021

    Maria Schuld and Francesco Petruccione.Machine learning with quantum computers, volume 676. Springer, 2021

    2021

Show all 39 references
  1. [9]

    Artificial intelligence computing at the quantum level.Data, 7(3):28, 2022

    Olawale Ayoade, Pablo Rivas, and Javier Orduz. Artificial intelligence computing at the quantum level.Data, 7(3):28, 2022

    2022

  2. [10]

    Robust online hamiltonian learning.New Journal of Physics, 14(10):103013, 2012

    Christopher E Granade, Christopher Ferrie, Nathan Wiebe, and David G Cory. Robust online hamiltonian learning.New Journal of Physics, 14(10):103013, 2012

    2012

  3. [11]

    Separability-entanglement classifier via machine learning.Physical Review A, 98(1):012315, 2018

    Sirui Lu, Shilin Huang, Keren Li, Jun Li, Jianxin Chen, Dawei Lu, Zhengfeng Ji, Yi Shen, Duanlu Zhou, and Bei Zeng. Separability-entanglement classifier via machine learning.Physical Review A, 98(1):012315, 2018

    2018

  4. [12]

    Entanglement verification with deep semisupervised machine learning

    Lifeng Zhang, Zhihua Chen, and Shao-Ming Fei. Entanglement verification with deep semisupervised machine learning. Physical Review A, 108(2):022427, 2023

    2023

  5. [13]

    Optimal entanglement witness of multipartite systems using support vector machine approach.arXiv preprint arXiv:2504.18163, 2025

    Mahmoud Mahdian and Zahra Mousavi. Optimal entanglement witness of multipartite systems using support vector machine approach.arXiv preprint arXiv:2504.18163, 2025

    2025 arXiv

  6. [14]

    Quantum machine learning.Nature, 549(7671):195–202, 2017

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Rebentrost, Nathan Wiebe, and Seth Lloyd. Quantum machine learning.Nature, 549(7671):195–202, 2017

    2017

  7. [15]

    Quantum information with continuous variables.Reviews of modern physics, 77(2):513–577, 2005

    Samuel L Braunstein and Peter Van Loock. Quantum information with continuous variables.Reviews of modern physics, 77(2):513–577, 2005

    2005

  8. [16]

    Gaussian quantum information.Reviews of Modern Physics, 84(2):621–669, 2012

    Christian Weedbrook, Stefano Pirandola, Ra´ ul Garc´ ıa-Patr´ on, Nicolas J Cerf, Timothy C Ralph, Jeffrey H Shapiro, and Seth Lloyd. Gaussian quantum information.Reviews of Modern Physics, 84(2):621–669, 2012

    2012

  9. [17]

    Experimentally realizable continuous-variable quantum neural networks.Physical Review A, 108(4):042414, 2023

    Shikha Bangar, Leanto Sunny, K¨ ubra Yeter-Aydeniz, and George Siopsis. Experimentally realizable continuous-variable quantum neural networks.Physical Review A, 108(4):042414, 2023

    2023

  10. [18]

    Continuous- variable quantum neural networks.Physical Review Research, 1(3):033063, 2019

    Nathan Killoran, Thomas R Bromley, Juan Miguel Arrazola, Maria Schuld, Nicol´ as Quesada, and Seth Lloyd. Continuous- variable quantum neural networks.Physical Review Research, 1(3):033063, 2019

    2019

  11. [19]

    Quantum computation over continuous variables.Physical Review Letters, 82(8):1784, 1999

    Seth Lloyd and Samuel L Braunstein. Quantum computation over continuous variables.Physical Review Letters, 82(8):1784, 1999

    1999

  12. [20]

    Optimal design for universal multiport interferometers.Optica, 3(12):1460–1465, 2016

    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, 2016

    2016

  13. [21]

    Springer Science & Business Media, 2004

    Matteo Paris and Jaroslav Rehacek.Quantum state estimation, volume 649. Springer Science & Business Media, 2004

    2004

  14. [22]

    Cambridge university press, 1997

    Ulf Leonhardt.Measuring the quantum state of light, volume 22. Cambridge university press, 1997

    1997

  15. [23]

    Quantum circuit learning.Physical Review A, 98(3):032309, 2018

    Kosuke Mitarai, Makoto Negoro, Masahiro Kitagawa, and Keisuke Fujii. Quantum circuit learning.Physical Review A, 98(3):032309, 2018

    2018

  16. [24]

    Evaluating analytic gradients on quantum hardware.Physical Review A, 99(3):032331, 2019

    Maria Schuld, Ville Bergholm, Christian Gogolin, Josh Izaac, and Nathan Killoran. Evaluating analytic gradients on quantum hardware.Physical Review A, 99(3):032331, 2019. 16

    2019

  17. [25]

    Classification with quantum neural networks on near term processors.arXiv preprint arXiv:1802.06002, 2018

    Edward Farhi and Hartmut Neven. Classification with quantum neural networks on near term processors.arXiv preprint arXiv:1802.06002, 2018

    2018 arXiv

  18. [26]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  19. [27]

    Multilayer feedforward networks are universal approximators

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators. Neural networks, 2(5):359–366, 1989

    1989

  20. [28]

    Strawberry fields: A software platform for photonic quantum computing.Quantum, 3:129, 2019

    Nathan Killoran, Josh Izaac, Nicol´ as Quesada, Ville Bergholm, Matthew Amy, and Christian Weedbrook. Strawberry fields: A software platform for photonic quantum computing.Quantum, 3:129, 2019

    2019

  21. [29]

    Tensorflow: Large-scale machine learning on heterogeneous distributed systems.arXiv preprint arXiv:1603.04467, 2016

    Mart´ ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al. Tensorflow: Large-scale machine learning on heterogeneous distributed systems.arXiv preprint arXiv:1603.04467, 2016. 17 Appen...

    2016 arXiv

  22. [30]

    This means there is a one-to-one and continuous correspondence between quantum states inXand classical vectors inP

    Homeomorphism via IC Measurement:Since the feature map gΘ∗ :X→R K is injective by the definition of an IC-POVM, and continuous by Lemma 1, it establishes a homeomorphism between the compact set of states X and its image, the compact set of classical feature vectors P =gΘ∗(X )....

  23. [31]

    In other words, the task of learning the quantum functional F is equivalent to learning the classical function ˜Fon the feature spaceP

    Induced Functional on Classical Space:For any arbitrary continuous functional we wish to learn, F :X→R , this homeomorphism guarantees the existence of a corresponding continuous function ˜F :P→R such that F (ρ) = ˜F (gΘ∗(ρ)). In other words, the task of learning the quantum f...

  24. [32]

    Universal Approximation on Feature Space:The classical feature space P is a compact subset of RK. The Universal Approximation Theorem for neural networks states that a standard MLP with at least one hidden layer and a non-polynomial activation function (like ReLU or sigmoid) c...

  25. [33]

    The parameters of all gates (rotations, beamsplitters, squeezers, displacements, Kerr) were initialized randomly by sampling from a normal distributionN(0,0.1)

    CV-QNN Architecture Details • Quantum Circuit:The variational ansatz consists of L = 2 layers for all experiments. The parameters of all gates (rotations, beamsplitters, squeezers, displacements, Kerr) were initialized randomly by sampling from a normal distributionN(0,0.1). •...

  26. [34]

    • Support Vector Machine (SVM):We used an SVM with a non-linear Radial Basis Function (RBF) kernel

    Classical Baseline Architectures and Tuning To ensure a fair and rigorous comparison, the classical baseline models were carefully selected and their hyperparam- eters were thoroughly tuned. • Support Vector Machine (SVM):We used an SVM with a non-linear Radial Basis Function ...

  27. [35]

    To provide a reliable measure of the uncertainty in our performance metrics, we computed 95% confidence intervals (CIs) using the **stratified bootstrap** method

    Stratified Bootstrap for Confidence Intervals Point estimates of accuracy can be misleading, especially with a finite test set. To provide a reliable measure of the uncertainty in our performance metrics, we computed 95% confidence intervals (CIs) using the **stratified bootst...

  28. [36]

    Let the test set beT={(x i,yi)}Ntest i=1

  29. [37]

    Forb= 1,...,B(where we useB= 1000 bootstrap resamples): (a) Create a new bootstrap sampleT b by drawingN test samples fromTwith replacement. (b) To maintain the class distribution of the original test set, we perform the sampling in a stratified manner: samples for the ’separa...

  30. [38]

    The collection of scores{A 1,...,A B}forms the bootstrap distribution of the accuracy

  31. [39]

    An observed performance gap between two models is considered statistically significant if their 95% CIs are non- overlapping

    The 95% confidence interval is then calculated by taking the 2.5th and 97.5th percentiles of this distribution. An observed performance gap between two models is considered statistically significant if their 95% CIs are non- overlapping. As seen in Table II, the CIs for the CV...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.