Pith. sign in

REVIEW 3 major objections 5 minor 59 references

Quantum Machine Learning Applied to the Sinking of the Titanic

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read Non-entangling quantum encoding tops entangling map on Titanic data

desk verdict Routine QML benchmark with a load-bearing robustness claim that rests on a single train/test split; the feature-map comparison is worth a look, the SVC 'collapse' is not established. read the letter →

arxiv 2509.00916 v1 pith:TIZQHLSX submitted 2025-08-31 quant-ph

classification quant-ph
keywords variationalquantumclassifierTitanicdatasetPauliexpansionfeaturemapsRealAmplitudesansatzCOBYLAoptimizersupportvectorbinaryclassificationhybridquantum-classical
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 the titanic3 survival dataset, hybrid quantum-classical variational classifiers built with the non-entangling ZFeatureMap consistently outperform those built with the entangling ZZFeatureMap, with accuracy and balanced accuracy saturating near 0.80 once the ansatz has 15 to 20 variational parameters. It also claims that the small quantum classifiers match a classical linear support vector classifier across most training sizes, and that the classical model collapses at 90% training data while the quantum models do not. If correct, the result supports the practical viability of shallow variational quantum classifiers for small classical binary classification tasks in the NISQ era.

What carries the argument

The models couple a Pauli expansion feature map with a variational ansatz. The ZFeatureMap encodes each feature into single-qubit phase gates with no entangling operations, while the ZZFeatureMap adds entangling cX gates and nonlinear phase terms. The RealAmplitudes ansatz applies alternating layers of Ry rotations and cX entangling gates; the rotation angles are the variational parameters optimized by the COBYLA gradient-free optimizer to minimize cross-entropy loss. An ideal statevector simulation supplies the measurement probabilities. The ZFeatureMap/RealAmplitudes combination is the mechanism that carries the argument, since the paper attributes the performance gap to the choice of enco

What would settle it

Run a repeated random-splitting experiment, for example 50 different 90/10 splits of titanic3, training the linear SVC and the Z20 quantum classifier on each split; if the SVC's Youden's index remains above 0.2 on most splits, the reported collapse at 90% training was an artifact of the single split. Similarly, the ZFeatureMap-versus-ZZFeatureMap ranking could be tested with repeated initializations and confidence intervals.

Watch

Extended reading notes

Core claim

The central finding is that a five-qubit variational classifier with a non-entangling first-order Pauli expansion feature map and a RealAmplitudes ansatz reaches accuracy and balanced accuracy near 0.80 on the titanic3 dataset, with performance saturating beyond 15 to 20 variational parameters. Identical models using the entangling ZZFeatureMap fluctuate around 0.60 to 0.75 and show low Youden's indices, indicating difficulty separating the classes. In a head-to-head comparison against a linear support vector classifier across training fractions from 10% to 90%, the Z20 quantum model performs comparably to the classical model at most sizes, but at 90% training the classical model's Youden's

Load-bearing premise

Each reported performance point comes from a single fixed train/test split, so the classical model's collapse at 90% training could be a property of that particular split rather than a systematic difference.

Editorial extensions

If this is right

  • For datasets with a small number of tabular features, non-entangling feature maps may be sufficient, or even preferable, to entangling maps in variational quantum classification.
  • Performance saturation at 15 to 20 variational parameters suggests shallow circuits can reach the practical ceiling for this task, which is relevant for trainability and noise resilience.
  • The reported absence of a high-training-size collapse in the quantum models, if confirmed, would indicate a robustness property worth studying in small-data regimes.
  • The paper provides a concrete template for benchmarking variational quantum classifiers against classical baselines using standard confusion-matrix metrics.

Reading between the lines

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

  • The claimed collapse of the classical SVC at 90% training rests on a single train/test split; repeated random splits or cross-validation could show the collapse is an artifact of that particular split, changing the robustness conclusion.
  • The superiority of ZFeatureMap over ZZFeatureMap may be dataset-specific: titanic3 has only five features and largely linear correlations, so adding entanglement may add unwanted expressiveness; on strongly nonlinear data the ranking could reverse.
  • A natural testable extension is to run the same comparison with multiple seeds and K-fold splits, reporting confidence intervals for accuracy, balanced accuracy, and Youden's index across training sizes.
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

3 major / 5 minor

Summary. The paper applies hybrid quantum-classical variational classifiers to the titanic3 dataset. Five features are encoded with either the non-entangling ZFeatureMap or the entangling ZZFeatureMap, followed by a RealAmplitudes ansatz with 10–50 variational parameters, trained by COBYLA on cross-entropy loss in an ideal statevector simulation. The authors report that ZFeatureMap models outperform ZZFeatureMap models and saturate near 0.80 accuracy/balanced accuracy beyond roughly 15–20 parameters. They then compare Z15 and Z20 with a classical linear SVC across training fractions and observe, at 90% training, a collapse of the SVC's Youden index to about zero, a failure mode they claim is absent in the quantum classifiers. The abstract interprets this as evidence of robustness and viability of variational quantum classifiers.

Significance. If the reported results were properly supported, the paper would provide a valuable empirical counterpoint to the common assumption that entanglement in the feature map is needed for quantum advantage on classical data, and it would document a concrete robustness behavior of small variational classifiers. The study is transparent in its use of public libraries, standard metrics, and a well-known dataset; there is no circularity in the evaluation, and the Z20 ten-run experiment is a useful step. The main significance, however, rests on two quantitative claims—the Z-vs-ZZ comparison and the SVC collapse—neither of which is accompanied by a statistical assessment of split variability. The paper is therefore a useful exploratory study but not yet a definitive demonstration of the claimed robustness advantage.

major comments (3)
  1. [§3.2, Figures 7 and 8] The central claim that the classical SVC 'collapses' at 90% training is based on a single train/test split for each training fraction. At 90% training the test set has roughly 131 instances, with about 50 survivors and 81 non-survivors; TPR and TNR each have standard errors of order 0.05–0.07, and Youden's J has a standard error of about 0.09. A single split can therefore move J from a healthy value to near zero. The authors perform 10 runs for Z20 but do not resample the data split. Because this collapse is the only point where the quantum models beat the classical baseline and is highlighted in the abstract, it must be supported by repeated random splits (or cross-validation), with means, standard deviations, or confidence intervals. The SVC's default decision threshold and class weighting should also be examined; the collapse may be an artifact of an uncalibrated threshold on an imbal
  2. [§3.1, Figures 4 and 5] The claim that ZFeatureMap models 'consistently outperformed' ZZFeatureMap models is based on point estimates from one training/test split and one optimizer run per parameter count, except for the Z20 ten-run check. The visible differences in accuracy, balanced accuracy, and Youden's index are not accompanied by any measure of variance or significance. Since both the data split and the optimizer initialization are nondeterministic, the observed gaps could be within run-to-run or split-to-split fluctuation. This is load-bearing for the paper's first main conclusion and needs at least repeated randomized runs with standard deviations, or a paired comparison over multiple splits.
  3. [§2, paragraph after Eq. (3)] All models are trained with a fixed 150 epochs, but this value is inferred from a single training curve for Z35 (Fig. 3). Larger ansatze (Z40–Z50, ZZ40–ZZ50) are likely to require more optimization steps to converge, and no convergence check is reported for them. The observed saturation (or lack of improvement) beyond 15–20 parameters could therefore be an artifact of under-training the larger models rather than a property of the model class. The authors should verify convergence for each parameter count or justify that 150 epochs is uniformly sufficient.
minor comments (5)
  1. [§3, 'contingency table' paragraph] Typographical issues: 'true negative (TP)' should be 'true negative (TN)', and 'perisehd' should be 'perished'. These should be corrected.
  2. [References] Reference [59] has malformed URL: 'htpps://doi.or/10.1007/BF00994018' should be corrected to 'https://doi.org/10.1007/BF00994018'.
  3. [General] The manuscript does not state whether code or data are available, nor the exact random seeds and software versions. This limits reproducibility; the authors should add a data/code availability statement.
  4. [§2, Table 1] The relationship between the number of variational parameters and the number of qubits/repetitions of the RealAmplitudes ansatz is implicit. A short explanation or a column with circuit depth would help readers interpret the 10-to-50 parameter range.
  5. [§3.2] For the SVC comparison, the authors should specify all non-default hyperparameters, including the class_weight setting and the decision threshold. This matters because the 90% collapse may be threshold-induced.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the reported metrics are genuine held-out predictions from trained models.

full rationale

The paper's chain is empirical, not derivational. Variational quantum classifiers are trained with COBYLA on cross-entropy loss using training splits, then evaluated on held-out test splits; no fitted parameter is repackaged as a prediction and no quantity used as input is also the claimed output. The comparison between ZFeatureMap and ZZFeatureMap is a direct experimental observation on the same train/test protocol. The later benchmark against a classical SVC uses the same held-out protocol across training sizes, so the reported accuracy, balanced accuracy, and Youden's index are genuine out-of-sample measurements. There is no load-bearing self-citation: the cited works are standard references for Qiskit, COBYLA, scikit-learn, and dataset sources, none of which supply a forced conclusion. The Z20 repeated-instance study addresses quantum nondeterminism, and although the single-split SVC collapse at 90% training is a statistical robustness concern, that is a question of experimental design, not circular reasoning. No equation is defined in terms of another result, and no known empirical pattern is merely renamed. Therefore no circularity is present.

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

The paper introduces no new entities or mathematical derivation. The listed free parameters are experimental hyperparameters and a data imputation statistic. The axioms are the implicit assumptions behind the empirical setup: ideal simulation, feature sufficiency, split representativeness, and optimizer adequacy.

free parameters (4)
  • Training epochs = 150
    Set by inspecting convergence of one model (Z35) at 70% data, Section 2. This hand-picked value could affect all model results.
  • Mean imputation for age = mean of non-missing ages
    Missing ages (263 records) replaced with the mean age, Section 2. The mean is a data-derived statistic.
  • Number of variational parameters scanned = 10, 15, 20, 25, 30, 35, 40, 45, 50
    Chosen to test saturation; the specific values are arbitrary but cover a range.
  • Training fraction = 0.70 for main comparison; 0.10-0.90 in increments for SVC benchmark
    Training sizes chosen by the authors; the 0.90 case is where the collapse is observed.
assumptions (4)
  • domain assumption Ideal statevector simulation is an adequate proxy for NISQ hardware
    All results come from an ideal simulator; noise, decoherence, and shot noise on real hardware could change relative performance. Section 2 mentions the Statevector simulator.
  • domain assumption The cleaned set of five features contains sufficient predictive signal
    Features were selected and missing values imputed by the authors; the assumption is that survival is learnable from these features. Section 2.
  • domain assumption Single random train/test splits are representative
    No repeated splits or cross-validation are reported; the comparison assumes the chosen splits are not unusual. Sections 3.1 and 3.2.
  • domain assumption COBYLA converges to an acceptable local optimum for all models
    The optimizer was run with default settings and 150 epochs; no check across random initializations except for Z20. Section 2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Machine Learning Applied to the Sinking of the Titanic." pith.science (2026). https://pith.science/paper/TIZQHLSX

@misc{pith2026250900916,
  author       = {Pith},
  title        = {Pith review of: Quantum Machine Learning Applied to the Sinking of the Titanic},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TIZQHLSX}},
  note         = {Machine review of arXiv:2509.00916}
}
read the original abstract

This work investigates the performance of hybrid quantum-classical variational classifiers applied to a supervised learning task involving the titanic3 dataset. Quantum models were constructed using Pauli entangling and non-entangling expansion-based feature maps and the RealAmplitudes ansatz with up to 50 variational parameters. Model training employed the COBYLA gradient-free optimizer to minimize the cross-entropy loss, within an ideal statevector simulation framework. Comparative performance analysis reveals that the models based on the non-entangling feature map consistently outperformed the models based on the entangling features maps, achieving saturation of classification metrics (accuracy, balanced accuracy, and Youden's index) beyond 15 to 20 parameters. Further, two quantum models were benchmarked against a classical Support Vector Classifier (SVC). While both approaches yielded similar predictive performance across multiple training sizes, the classical model exhibited a performance collapse when trained with 90% of the dataset, a failure mode absent in the quantum classifiers. These results underscore the robustness and viability of variational quantum classifiers for binary classification tasks on classical datasets in the NISQ era.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 26 canonical work pages

  1. [1]

    Wiesner, Conjugate coding

    S. Wiesner, Conjugate coding. ACM SIGACT News 15(1), 78-88 (1983). https: //doi.org/10.1145/1008908.1008920

  2. [2]

    C. H. Bennett, G. Brassard, Quantum cryptography: Public key distribution and coin tossing. Theor. Comput. Sci. 560, 7-11 (2014). https://doi.org/10.1016/j.tcs. 2014.05.025

  3. [3]

    A. K. Ekert, Quantum cryptography based on Bell’s theorem. Phys. Rev. Lett. 67, 661-663 (1991). https://doi.org/10.1103/PhysRevLett.67.661

  4. [4]

    Park, The concept of transition in quantum mechanics

    J. Park, The concept of transition in quantum mechanics. Found. Phys. 1(1), 23-33 (1970). https://doi.org/10.1007/BF00708652

  5. [5]

    W. K. Wootters and W. H. Zurek, A single quantum cannot be cloned. Nature 299, 802-803 (1982). https://doi.org/10.1038/299802a0

  6. [6]

    Dieks, Communication by EPR devices

    D. Dieks, Communication by EPR devices. Phys. Lett. A 92, 271-272 (1982). https://doi.org/10.1016/0375-9601(82)90084-6 16

  7. [7]

    P. W. Milonni and M. L. Hardies, Photons cannot always be replicated. Phys. Lett. A 92, 321-322 (1982). https://doi.org/10.1016/0375-9601(82)90899-4

  8. [8]

    R. P. Feynman, Simulating physics with computers. Int. J. Theor. Phys. 21, 467- 488 (1982). https://doi.org/10.1007/BF02650179

Show all 59 references
  1. [9]

    R. P. Feynman, Quantum mechanical computers. Found. Phys.16, 507-531 (1986). https://doi.org/10.1007/BF01886518

  2. [10]

    Benioff, The computer as a physical system: A microscopic quantum mechan- ical Hamiltonian model of computers as represented by Turing machines

    P. Benioff, The computer as a physical system: A microscopic quantum mechan- ical Hamiltonian model of computers as represented by Turing machines. J. Stat. Phys. 22, 563-591 (1980). https://doi.org/10.1007/BF01011339

  3. [11]

    P. A. Benioff, Quantum mechanical Hamiltonian models of discrete processes that erase their own histories: Application to Turing Machines. Int. J. Theor. Phys. 21, 177-201 (1982). https://doi.org/10.1007/BF01857725

  4. [12]

    Benioff, Quantum mechanical Hamiltonian models of Turing machines

    P. Benioff, Quantum mechanical Hamiltonian models of Turing machines. J. Stat. Phys. 29, 515-546 (1982). https://doi.org/10.1007/BF01342185

  5. [13]

    P. Benioff. Quantum mechanical models of Turing machines that dissipate no energy. Phys. Rev. Lett. 48, 1581-1585 (1982). https://doi.org/10.1103/ PhysRevLett.48.1581

  6. [14]

    Deutsch, Quantum theory, the Church-Turing principle and the universal quantum computer

    D. Deutsch, Quantum theory, the Church-Turing principle and the universal quantum computer. Proc. R. Soc. A 400, 97-117 (1985). http://doi.org/10.1098/ rspa.1985.0070

  7. [15]

    Deutsch and R

    D. Deutsch and R. Jozsa, Rapid solutions of problems by quantum computation. Proc. R. Soc. A 439, 553-558 (1992). https://doi.org/10.1098/rspa.1992.0167 17

  8. [16]

    C. H. Bennett and S. J. Wiesner, Communication via one- and two-particle oper- ators on Einstein-Podolsky-Rosen states. Phys. Rev. Lett 69, 2881-2884 (1992). https://doi.org/10.1103/PhysRevLett.69.2881

  9. [17]

    Bernstein, U

    E. Bernstein, U. Vazirani, Quantum complexity theory, Proceedings of the 25th annual ACM Symposium on Theory of Computing, 11-20 (1993). https://doi.org/ 10.1145/167088.167097

  10. [18]

    Bernstein and U

    E. Bernstein and U. Vazirani, Quantum complexity theory. SIAM J. Computing 26, 1411-1473 (1997). https://doi.org/10.1137/S0097539796300921

  11. [19]

    C. H. Bennett, G. Brassard, C. Cr´ epeau, R. Jozsa, A. Peres, and W. K. Wootters, Teleporting an unknown quantum state via dual classical and Einstein-Podolsky- Rosen channels. Phys. Rev. Lett. 70, 1895-1899 (1993). https://doi.org/10.1103/ PhysRevLett.70.1895

  12. [21]

    D. R. Simon, On the power of quantum computation. SIAM J. Computing 26, 1474-1483 (1997). https://doi.org/10.1137/S0097539796298637

  13. [22]

    P. W. Shor, Algorithms for quantum computation: Discrete logarithms and fac- toring. Proceedings of the 35th annual Symposium on Foundations of Computer Science, 124-134 (1994). https://doi.org/10.1109/sfcs.1994.365700

  14. [23]

    P. W. Shor, Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer. SIAM J. Computing 26, 1484-1509 (1997). https://doi.org/10.1137/S0097539795293172 18

  15. [24]

    Coppersmith, An approximate Fourier transform useful in quantum factor- ing

    D. Coppersmith, An approximate Fourier transform useful in quantum factor- ing. Preprint at https://arxiv.org/abs/quant-ph/0201067 (1994). https://doi.org/ 10.48550/arXiv.quant-ph/0201067

  16. [25]

    A. Yu. Kitaev, Quantum measurements and the Abelian stabilizer problem. Preprint at https://arxiv.org/abs/quant-ph/9511026 (1995). https://doi.org/10. 48550/arXiv.quant-ph/9511026

  17. [26]

    L. K. Grover, A fast quantum mechanical algorithm for database search. Proceed- ings of the twenty-eighth annual ACM symposium on theory of computing, 212-219 (1996). https://doi.org/10.1145/237814.237866

  18. [27]

    L. K. Grover, Quantum mechanics helps in searching for a needle in a haystack. Phys. Rev. Lett. 79, 325-328 (1997). https://doi.org/10.1103/PhysRevLett.79.325

  19. [28]

    Brassard, P

    G. Brassard, P. Høyer, M. Mosca, and A. Tapp, Quantum amplitude amplification and estimation. In Quantum computation and information (eds. S. J. Lomonaco, Jr. and H. E. Brandt), 53-74 (American Mathematical Society, 2002). https://doi. org/10.1090/conm/305

  20. [29]

    A. W. Harrow, A. Hassidim, and S. Lloyd, Quantum algorithm for solving linear systems of equations. Phys. Rev. Lett.103, 150502 (2009). https://doi.org/10.1103/ PhysRevLett.103.150502

  21. [30]

    A ¨ ımeur, G

    E. A ¨ ımeur, G. Brassard, and S. Gambs, Machine learning in a quantum world.In Advances in Artificial Intelligence (eds. L. Lamontagne and M. Marchand), 431-442 (Springer, 2006). https://doi.org/10.1007/11766247 37

  22. [31]

    Lloyd, M

    S. Lloyd, M. Mohseni, and P. Rebentrost, Quantum algorithms for supervised and unsupervised machine learning. Preprint at https://arxiv.org/abs/1307.0411 (2013). https://doi.org/10.48550/arXiv.1307.0411 19

  23. [32]

    Wittek, Quantum Machine Learning: What Quantum Computing Means to Data Mining (Academic Press, 2014)

    P. Wittek, Quantum Machine Learning: What Quantum Computing Means to Data Mining (Academic Press, 2014)

  24. [33]

    Schuld, I

    M. Schuld, I. Sinayskiy, and F. Petruccione, An introduction to quantum machine learning. Contemp. Phys. 56, 172-185 (2015). https://doi.org/10.1080/00107514. 2014.964942

  25. [34]

    Biamonte, P

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, Quantum machine learning. Nature 549, 195-202 (2017). https://doi.org/10.1038/ nature23474

  26. [35]

    Dunjko, and H

    V. Dunjko, and H. J. Briegel, Machine learning & artificial intelligence in the quantum domain: a review of recent progress. Rep. Prog. Phys. 81, 074001 (2018). https://doi.org/10.1088/1361-6633/aab406

  27. [36]

    Benedetti, E

    M. Benedetti, E. Lloyd, S. Sack, and M. Fiorentini, Parameterized quantum cir- cuits as machine learning models. Quant. Sci. Technol. 4, 043001 (2019). https: //doi.org/10.1088/2058-9565/ab4eb5

  28. [37]

    Havl ´ ıˇ cek, A

    V. Havl ´ ıˇ cek, A. D. C´ orcoles, K. Temme, A. W. Harrow, A. Kandala, J. M. Chow, and J. M. Gambetta, Supervised learning with quantum-enhanced feature spaces. Nature 567, 209-212 (2019). https://doi.org/10.1038/s41586-019-0980-2

  29. [38]

    Javadi-Abhari, M

    A. Javadi-Abhari, M. Treinish, K. Krsulich, C. J. Wood, J. Lishman, J. Gacon, S. Martiel, P. Nation, L. S. Bishop, A. W. Cross, B. R. Johnson, and J. M. Gambetta, (2024, May 15). Quantum computing with Qiskit. Preprint at arxiv.org/abs/2405. 08810 (2024). https://doi.org/10.48...

  30. [39]

    Schuld, F

    M. Schuld, F. Petruccione, Machine Learning with Quantum Computers (Springer, 2021) 2nd edition. 20

  31. [40]

    Cerezo, A

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, and P. J. Coles, Variational quantum algorithms. Nat. Rev. Phys. 3, 625-644 (2021). https://doi.org/10.1038/ s42254-021-00348-9

  32. [41]

    Cerezo, G

    M. Cerezo, G. Verdon, H.-Y. Huang, L. Cincio, and P. J. Coles, Challenges and opportunities in quantum machine learning. Nat. Comput. Sci. 2, 567-576 (2022). https://doi.org/10.1038/s43588-022-00311-3

  33. [42]

    M. P. Deisenroth, A. A. Faisal, and C. S. Ong,Mathematics for Machine Learning (Cambridge University Press, 2020)

  34. [43]

    Sevilla, L

    J. Sevilla, L. Heim, A. Ho, T. Besiroglu, M. Hobbhahn, and P. Villalobos, Com- pute Trends Across Three Eras of Machine Learning, 2022 International Joint Conference on Neural Networks, 1-8 (2022). https://doi.org/10.1109/IJCNN55064. 2022.9891914

  35. [44]

    Preskill, Quantum computing in the NISQ era and beyond

    J. Preskill, Quantum computing in the NISQ era and beyond. Quantum 2, 79 (2018). https://doi.org/10.22331/q-2018-08-06-79

  36. [45]

    Schuld, V

    M. Schuld, V. Bergholm, C. Gogolin,J. Izaac, and N. Killoran, Evaluating analytic gradients on quantum hardware. Phys. Rev. A 99, 032331 (2019). https://doi.org/ 10.1103/PhysRevA.99.032331

  37. [46]

    Larocca, S

    M. Larocca, S. Thanasilp, S. Wang, K. Sharma, J. Biamonte, P. J. Coles, L. Cincio, J. R. McClean, Z. Holmes, and M. Cerezo, Barren plateaus in variational quantum computing. Nat. Rev. Phys. 7, 174-189 (2025). https://doi.org/10.1038/ s42254-025-00813-9

  38. [47]

    Arrasmith, M

    A. Arrasmith, M. Cerezo, P. Czarnik, L. Cincio, and P. J. Coles, Effect of barren plateaus on gradient-free optimization. Quantum 5, 558 (2021). https://doi.org/10. 21 22331/q-2021-10-05-558

  39. [48]

    E. R. Anschuet and B. T. Kiani, Quantum variational algorithms are swamped with traps. Nat. Commun. 13, 7760 (2022). https://doi.org/10.1038/ s41467-022-35364-5

  40. [49]

    F. E. Harrell Jr, Regression Modeling Strategies with Applications to Linear Mod- els, Logistic Regression, and Survival Analysis (Springer, 2001). https://doi.org/10. 1007/978-3-319-19425-7

  41. [50]

    F. E. Harrell Jr. and T. Cason, Titanic3 dataset. Dataset avaiable at https:// hbiostat.org/data (2002)

  42. [51]

    https://www.kaggle.com/c/ titanic/overview (2012)

    Kaggle, Titanic - machine learning from disaster. https://www.kaggle.com/c/ titanic/overview (2012)

  43. [52]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cour- napeau, M. Brucher, M. Perrot, and E. Duchesnay, Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12, ...

  44. [53]

    M. J. D. Powell, A direct search optimization method that models the objective and constraint functions by linear interpolation. In Advances in Optimization and Numerical Analysis (eds. S. Gomez and J.-P. Hennart), 51-67 (Kluwer, 1994). https: //doi.org/10.1007/978-94-015-8330-5 4

  45. [54]

    M. E. Sahin, E. Altamura, O. Wallis, S. P. Wood, A. Dekusar, D. A. Millar, T. Imamichi, A. Matsuo, and S. Mensa, Qiskit Machine Learning: an open-source library for quantum machine learning tasks at scale on quantum hardware and classical simulators. Preprint at https://arxiv....

  46. [55]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cour- napeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. van der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Larson, C. J. Carey, ˙I. Polat, Y. Fe...

  47. [56]

    Kullback and R

    S. Kullback and R. A. Leibler, On information and sufficiency. Ann. Math. Statist. 22, 79–86 (1951). https://doi.org/10.1214/aoms/1177729694

  48. [57]

    Sokolova and G

    M. Sokolova and G. Lapalme, A systematic analysis of performance measures for classification tasks. Inf. Process. Manag. 45, 427–437 (2009). https://doi.org/10. 1016/j.ipm.2009.03.002

  49. [58]

    Tharwat, Classification assessment methods

    A. Tharwat, Classification assessment methods. Appl. Comput. Inform. 17, 168– 19 (2021). https://doi.org/10.1016/j.aci.2018.08.003

  50. [59]

    Cortes and V

    C. Cortes and V. Vapnik, Support-vector networks. Machine Learning 20, 273– 297 (1995). htpps://doi.or/10.1007/BF00994018

  51. [60]

    S. L. Brunton and J. N. Kutz, Data-Driven Science and Engineering (Cambridge University Press, 2022) 2nd edition. 23

Pith tools

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