pith. sign in

arxiv: 2604.13608 · v1 · submitted 2026-04-15 · 💻 cs.LG · cs.AI

Design Space Exploration of Hybrid Quantum Neural Networks for Chronic Kidney Disease

Muhammad Kashif , Hanzalah Mohamed Siraj , Nouhaila Innan , Alberto Marchisio , Muhammad Shafique This is my paper

Pith reviewed 2026-05-10 13:51 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords hybrid quantum neural networkschronic kidney diseasedesign space explorationquantum machine learningdata encodingentanglement architecturesmedical diagnosisnear-term quantum computing
0
0 comments X

The pith

Compact hybrid quantum circuits with IQP encoding and ring entanglement deliver the best accuracy-robustness-efficiency trade-off for chronic kidney disease diagnosis.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper exhaustively benchmarks 625 hybrid quantum neural network models on a curated clinical dataset for chronic kidney disease diagnosis. It shows that design choices in data encoding, circuit architecture, measurement, and shots interact in non-obvious ways, so that high performance does not require large parameter counts or deep circuits. A reader would care because the results point to practical ways to run useful quantum machine learning on near-term hardware for a concrete medical task without excessive resources.

Core claim

The central claim is that compact architectures combined with appropriate encodings, such as IQP with Ring entanglement, achieve the best trade-off between accuracy, robustness, and efficiency among the 625 tested HQNNs; systematic 10-fold cross-validation evaluation reveals strong interactions between encoding schemes and entanglement architectures that outweigh raw model size or complexity.

What carries the argument

The design space exploration across five encoding schemes, five entanglement architectures, five measurement strategies, and five shot settings, which uncovers non-trivial encoding-architecture interactions that govern overall performance.

If this is right

  • Compact HQNN configurations can match or exceed more complex models across accuracy, AUC, F1, and composite scores for CKD without increasing parameter count.
  • Encoding-architecture pairs such as IQP with Ring entanglement optimize learning behavior more effectively than increasing circuit depth or shots.
  • Measurement strategies and shot counts exert measurable but secondary influence compared with the primary encoding and architecture choices.
  • The systematic benchmarking method supplies a reusable template for evaluating HQNN design spaces in other binary medical classification problems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the encoding-architecture interactions prove stable, similar compact designs could reduce the resource barrier for applying quantum ML to additional clinical tasks such as cancer staging or infection detection.
  • The results imply that near-term quantum devices may benefit more from careful classical-to-quantum data mapping than from scaling up circuit complexity in medical applications.
  • Validation on actual quantum hardware rather than classical simulators would test whether the simulated robustness of these compact models survives real device noise.

Load-bearing premise

The curated clinical dataset and the specific 625-model search space are representative enough that the observed encoding-architecture interactions will generalize to other medical datasets or real-world deployment conditions.

What would settle it

Repeating the identical 625-model search and 10-fold evaluation on an independent medical classification dataset, such as one for diabetes or heart disease, and finding that the top trade-off configurations shift away from compact IQP-ring models toward larger circuits would falsify the claimed generality.

Figures

Figures reproduced from arXiv: 2604.13608 by Alberto Marchisio, Hanzalah Mohamed Siraj, Muhammad Kashif, Muhammad Shafique, Nouhaila Innan.

Figure 1
Figure 1. Figure 1: Motivational Analysis highlighting the need for Design Space [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: HQNN design flow, from input to output. 2) Angle: Uses rotation gates to encode features as angles (hardware-friendly, one qubit per feature) [30]. 3) Basis: Maps discrete values to computational basis states (natural for binary/categorical features) [31]. 4) IQP encodes ‘n’ features in ‘n’ qubits using diagonal gates of an IQP circuit [32]. An IQP circuit is a quantum circuit of a block of Hadamards, foll… view at source ↗
Figure 3
Figure 3. Figure 3: An overview of our methodology which includes dataset comparability analysis, data preprocessing, and a grid search over hybrid model configurations [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Quantum Circuit Architectures with different entanglement structures [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Dataset 1: Summary of the 625-configuration HQNN design grid. (a) Factor-averaged GPS by encoding. (b) Factor-averaged GPS by entanglement topology (architecture). (c) Accuracy distribution by encoding. (d) Accuracy distribution by architecture. (e) Accuracy distribution by shot budget. (f) Accuracy distribution by measurement observable. (g) Specificity vs sensitivity across configurations. (h) MCC vs F1 … view at source ↗
Figure 6
Figure 6. Figure 6: Dataset 3: Summary of the 625-configuration HQNN design grid (same panel order as Dataset 1). Z/150 remains competitive for GPS1–2 and ranks second for both GPS3 and GPS4. Overall, the overlap pattern high￾lights Angle/Strong/Pauli-Y as the most stable choice, with IQP/Strong/Pauli-Z emerging as the primary alternative under balanced composites. B. Dataset 3 Analysis 1) Factor-averaged GPS trends (encoding… view at source ↗
Figure 7
Figure 7. Figure 7: (c–d) shows that accuracy is primarily shaped by encoding and architecture choices, with Strong/Star shifting the distribution upward relative to Alternating/Basic/Ring. The encoding-wise box plot indicates that competitive performance is concentrated in a subset of encodings, whereas the rest exhibit lower medians and/or larger dispersion. Measurement choice further modulates robustness and reveals notice… view at source ↗
read the original abstract

Hybrid Quantum Neural Networks (HQNNs) have recently emerged as a promising paradigm for near-term quantum machine learning. However, their practical performance strongly depends on design choices such as classical-to-quantum data encoding, quantum circuit architecture, measurement strategy and shots. In this paper, we present a comprehensive design space exploration of HQNNs for Chronic Kidney Disease (CKD) diagnosis. Using a carefully curated and preprocessed clinical dataset, we benchmark 625 different HQNN models obtained by combining five encoding schemes, five entanglement architectures, five measurement strategies, and five different shot settings. To ensure fair and robust evaluation, all models are trained using 10-fold stratified cross-validation and assessed on a test set using a comprehensive set of metrics, including accuracy, area under the curve (AUC), F1-score, and a composite performance score. Our results reveal strong and non-trivial interactions between encoding choices and circuit architectures, showing that high performance does not necessarily require large parameter counts or complex circuits. In particular, we find that compact architectures combined with appropriate encodings (e.g., IQP with Ring entanglement) can achieve the best trade-off between accuracy, robustness, and efficiency. Beyond absolute performance analysis, we also provide actionable insights into how different design dimensions influence learning behavior in HQNNs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The manuscript performs a systematic design-space exploration of 625 hybrid quantum neural networks for chronic kidney disease classification. Five encoding schemes, five entanglement architectures, five measurement strategies, and five shot counts are exhaustively combined and evaluated on a single curated clinical dataset using 10-fold stratified cross-validation together with held-out test metrics (accuracy, AUC, F1-score, and a composite score). The central empirical finding is that compact circuits paired with particular encodings (e.g., IQP with Ring entanglement) achieve the strongest accuracy–robustness–efficiency trade-off, and that non-trivial interactions exist between encoding and architecture choices.

Significance. If the reported interactions are reproducible, the study supplies concrete, quantitative guidance for near-term HQNN practitioners working on tabular medical data: high performance does not require large parameter counts or complex circuits. The scale of the enumerated search and the multi-metric evaluation protocol are positive features that could inform subsequent benchmarking efforts.

major comments (1)
  1. Abstract and §5 (Discussion): the claim that the observed encoding–architecture interactions yield 'actionable insights' for HQNN design in medical tabular tasks is load-bearing for the paper's broader contribution. This claim rests on a single preprocessed CKD cohort; no sensitivity analysis to changes in feature correlations, missing-value patterns, class balance, or input distribution is presented, so it is unclear whether the superiority of compact IQP-Ring models would persist on other medical datasets.
minor comments (3)
  1. §3 (Methods): the precise definition and weighting of the composite performance score should be stated explicitly rather than referenced only by name, to allow readers to interpret the ranking of the 625 models.
  2. Figures 3–5: error bars or standard deviations derived from the 10-fold CV folds are not shown; adding them would clarify whether the reported differences between top models are statistically distinguishable.
  3. §4 (Results): the preprocessing pipeline (missing-value imputation, feature scaling, and any class-balancing steps) is described at a high level; expanding this section with exact parameters would improve reproducibility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our design-space exploration of hybrid quantum neural networks. We address the major comment point by point below, with a commitment to revise the manuscript accordingly.

read point-by-point responses

  1. Referee: Abstract and §5 (Discussion): the claim that the observed encoding–architecture interactions yield 'actionable insights' for HQNN design in medical tabular tasks is load-bearing for the paper's broader contribution. This claim rests on a single preprocessed CKD cohort; no sensitivity analysis to changes in feature correlations, missing-value patterns, class balance, or input distribution is presented, so it is unclear whether the superiority of compact IQP-Ring models would persist on other medical datasets.

    Authors: We agree that the phrasing 'actionable insights' for HQNN design in medical tabular tasks is too broad given that all experiments use a single curated CKD dataset. No sensitivity analyses were performed with respect to feature correlations, missing-value patterns, class balance, or shifts in input distribution, and therefore we cannot claim that the superiority of compact IQP-Ring configurations would hold on other medical datasets. The empirical interactions we report are reproducible within this cohort and evaluation protocol, but their transferability remains an open question. To correct this, we will revise the abstract and §5 to replace the claim of general 'actionable insights' with a more precise statement that the observed encoding–architecture interactions provide concrete guidance for practitioners working on similar tabular clinical data, while explicitly noting the single-dataset limitation. We will also add a short limitations paragraph in §5 that acknowledges the absence of cross-dataset validation and recommends future benchmarking on additional medical tabular datasets. These textual changes will be made without altering the reported experimental results or methodology. revision: partial

Circularity Check

0 steps flagged

No circularity: purely empirical grid search with direct metric reporting

full rationale

The paper performs exhaustive enumeration of 625 HQNN configurations (5 encodings × 5 architectures × 5 measurements × 5 shot counts) on one preprocessed CKD dataset, trains each via 10-fold stratified CV, and reports observed accuracy/AUC/F1/composite scores. No equations, fitted parameters, or first-principles derivations are presented as predictions. No self-citation chain is invoked to justify uniqueness or to substitute for external validation. All central claims (e.g., compact IQP-Ring models achieving best trade-off) are direct measurements on the tested models and do not reduce to the inputs by construction. This is standard empirical benchmarking with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Only the abstract is available, so the ledger reflects standard assumptions in applied machine-learning papers rather than explicit statements in the manuscript. No free parameters or invented entities are described.

axioms (2)
  • domain assumption The curated and preprocessed clinical dataset is representative of real-world CKD cases and free of selection bias
    The abstract states a carefully curated dataset is used but provides no details on source, size, or validation against external populations.
  • domain assumption 10-fold stratified cross-validation plus the chosen metrics sufficiently capture model robustness for medical diagnosis
    Standard ML practice is invoked without discussion of potential distribution shift or clinical utility beyond the reported scores.

pith-pipeline@v0.9.0 · 5541 in / 1483 out tokens · 26916 ms · 2026-05-10T13:51:10.176576+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Towards Fair Benchmarking of Quantum Transfer Learning for Visual Classification

    quant-ph 2026-05 unverdicted novelty 6.0

    Introduces a unified benchmarking methodology for quantum transfer learning in visual classification tasks, finding that no single method dominates and performance varies with dataset, encoding, and circuit design.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    An introduction to quantum machine learning,

    M. Schuldet al., “An introduction to quantum machine learning,” Contemporary Physics, 2015

    work page 2015

  2. [2]

    Design space exploration of hybrid quantum classical neural networks,

    M. Kashif and S. Al-Kuwari, “Design space exploration of hybrid quantum classical neural networks,”Electronics, 2021

    work page 2021

  3. [3]

    A survey on quantum machine learning: Current trends, challenges, opportunities, and the road ahead,

    K. Zamanet al., “A survey on quantum machine learning: Current trends, challenges, opportunities, and the road ahead,”arXiv:2310.10315, 2023

    work page arXiv 2023

  4. [4]

    Financial fraud detection using quantum graph neural networks,

    N. Innan, A. Sawaikaet al., “Financial fraud detection using quantum graph neural networks,”Quantum Mach. Intell., 2024

    work page 2024

  5. [5]

    Qnn-vrcs: A quantum neural network for vehicle road cooperation systems,

    N. Innanet al., “Qnn-vrcs: A quantum neural network for vehicle road cooperation systems,”IEEE Trans. Intell. Transp. Syst., 2025

    work page 2025

  6. [6]

    Quiet-sr: Quantum image enhancement transformer for single image super-resolution,

    S. Duttaet al., “Quiet-sr: Quantum image enhancement transformer for single image super-resolution,”arXiv:2503.08759, 2025

    work page arXiv 2025

  7. [7]

    Position paper: Quantum neural networks - a paradigm shift in ai or a theoretical promise?

    M. Kashif and M. Shafique, “Position paper: Quantum neural networks - a paradigm shift in ai or a theoretical promise?” inIJCNN, 2025

    work page 2025

  8. [8]

    Next-generation quantum neural networks: Enhancing efficiency, security, and privacy,

    N. Innanet al., “Next-generation quantum neural networks: Enhancing efficiency, security, and privacy,” inIOLTS, 2025

    work page 2025

  9. [9]

    Deep quanvolutional neural networks with enhanced trainability and gradient propagation,

    M. Kashif and M. Shafique, “Deep quanvolutional neural networks with enhanced trainability and gradient propagation,”Scientific Reports, 2025

    work page 2025

  10. [10]

    Demonstrating quantum advantage in hybrid quantum neural networks for model capacity,

    M. Kashif and S. Al-Kuwari, “Demonstrating quantum advantage in hybrid quantum neural networks for model capacity,” inICRC, 2022

    work page 2022

  11. [11]

    Computational advantage in hybrid quantum neural networks: Myth or reality?

    M. Kashifet al., “Computational advantage in hybrid quantum neural networks: Myth or reality?” inDAC, 2025

    work page 2025

  12. [12]

    Barren plateaus in quantum neural network training landscapes,

    J. R. McCleanet al., “Barren plateaus in quantum neural network training landscapes,”Nature communications, 2018

    work page 2018

  13. [13]

    Resqnets: a residual approach for mitigating barren plateaus in quantum neural networks,

    M. Kashif and S. Al-Kuwari, “Resqnets: a residual approach for mitigating barren plateaus in quantum neural networks,”EPJ Quantum Technology, 2024

    work page 2024

  14. [14]

    Alleviating barren plateaus in parameterized quantum machine learning circuits: Investigating advanced parameter initialization strategies,

    M. Kashifet al., “Alleviating barren plateaus in parameterized quantum machine learning circuits: Investigating advanced parameter initialization strategies,” inDATE, 2024

    work page 2024

  15. [15]

    Nrqnn: The role of observable selection in noise-resilient quantum neural networks,

    M. Kashif and M. Shafique, “Nrqnn: The role of observable selection in noise-resilient quantum neural networks,” inWorld Congress in Computer Science, Computer Engineering & Applied Computing, 2024

    work page 2024

  16. [16]

    A comparative analysis and noise robustness evaluation in quantum neural networks,

    T. Ahmedet al., “A comparative analysis and noise robustness evaluation in quantum neural networks,”Scientific Reports, 2025

    work page 2025

  17. [17]

    Hqnet: Harnessing quantum noise for effective training of quantum neural networks in nisq era,

    M. Kashif and M. Shafique, “Hqnet: Harnessing quantum noise for effective training of quantum neural networks in nisq era,” inQAI, 2025

    work page 2025

  18. [18]

    Noisy hqnns: A comprehensive analysis of noise robustness in hybrid quantum neural networks,

    T. Ahmedet al., “Noisy hqnns: A comprehensive analysis of noise robustness in hybrid quantum neural networks,” inIJCNN, 2025

    work page 2025

  19. [19]

    Studying the impact of quantum-specific hyperparam- eters on hybrid quantum-classical neural networks,

    K. Zamanet al., “Studying the impact of quantum-specific hyperparam- eters on hybrid quantum-classical neural networks,” inWorld Congress in Computer Science, Computer Engineering & Applied Computing. Springer, 2024, pp. 132–149

    work page 2024

  20. [20]

    Faqnas: Flops- aware hybrid quantum neural architecture search using genetic algorithm,

    M. Kashifet al., “Faqnas: Flops-aware hybrid quantum neural architecture search using genetic algorithm,”arXiv:2511.10062, 2026

    work page arXiv 2026

  21. [21]

    Financial fraud detection: a comparative study of quantum machine learning models,

    N. Innanet al., “Financial fraud detection: a comparative study of quantum machine learning models,”Int. J. Quantum Inf., 2024

    work page 2024

  22. [22]

    Comparative performance analysis of quantum machine learning architectures for credit card fraud detection,

    M. El Alamiet al., “Comparative performance analysis of quantum machine learning architectures for credit card fraud detection,”Appl. Intell., 2026

    work page 2026

  23. [23]

    Effect of data encoding on the expressive power of variational quantum-machine-learning models,

    M. Schuldet al., “Effect of data encoding on the expressive power of variational quantum-machine-learning models,”Phys. Rev. A, 2021

    work page 2021

  24. [24]

    Connecting ansatz expressibility to gradient magnitudes and barren plateaus,

    Z. Holmeset al., “Connecting ansatz expressibility to gradient magnitudes and barren plateaus,”PRX Quantum, 2022

    work page 2022

  25. [25]

    The unified effect of data encoding, ansatz expressibility and entanglement on the trainability of hqnns,

    M. Kashif and S. Al-Kuwari, “The unified effect of data encoding, ansatz expressibility and entanglement on the trainability of hqnns,”IJPEDS, 2023

    work page 2023

  26. [26]

    Cutting is all you need: Execution of large-scale quantum neural networks on limited-qubit devices,

    A. Marchisioet al., “Cutting is all you need: Execution of large-scale quantum neural networks on limited-qubit devices,” inQAI, 2025

    work page 2025

  27. [27]

    t-distributed stochastic neighbor embedding,

    M. A. Anvariet al., “t-distributed stochastic neighbor embedding,” in Dimensionality Reduction in Machine Learning, 2025

    work page 2025

  28. [28]

    The impact of cost function globality and locality in hybrid quantum neural networks on nisq devices,

    M. Kashif and S. Al-Kuwari, “The impact of cost function globality and locality in hybrid quantum neural networks on nisq devices,”Machine Learning: Science and Technology, 2023

    work page 2023

  29. [29]

    Xanadu- Pennylane

    (2026) Pennylane documentation: Amplitude encoding. Xanadu- Pennylane. Accessed: January 2026. [Online]. Available: https://docs. pennylane.ai/en/stable/code/api/pennylane.AmplitudeEmbedding.html

    work page 2026

  30. [30]

    Xanadu-Pennylane

    (2026) Pennylane documentation: Angle encoding. Xanadu-Pennylane. Accessed: January 2026. [Online]. Available: https://docs.pennylane.ai/ en/stable/code/api/pennylane.AngleEmbedding.html

    work page 2026

  31. [31]

    Xanadu-Pennylane

    (2026) Pennylane documentation: Basis encoding. Xanadu-Pennylane. Accessed: January 2026. [Online]. Available: https://docs.pennylane.ai/ en/stable/code/api/pennylane.BasisEmbedding.html

    work page 2026

  32. [32]

    Xanadu-Pennylane

    (2026) Pennylane documentation: Iqpe encoding. Xanadu-Pennylane. Accessed: January 2026. [Online]. Available: https://docs.pennylane.ai/ en/stable/code/api/pennylane.IQPEmbedding.html

    work page 2026

  33. [33]

    Beyond bits: A review of quantum embedding techniques for efficient information processing,

    M. A. Khanet al., “Beyond bits: A review of quantum embedding techniques for efficient information processing,”IEEE Access, 2024

    work page 2024

  34. [34]

    Shot optimization in quantum machine learning architectures to accelerate training,

    K. Phalak and S. Ghosh, “Shot optimization in quantum machine learning architectures to accelerate training,”IEEE Access, 2023

    work page 2023

  35. [35]

    Shots and variance on noisy quantum circuits,

    M. Seksaria and A. Prabhakar, “Shots and variance on noisy quantum circuits,”arXiv:2501.03194, 2025

    work page arXiv 2025

  36. [36]

    Nearly optimal quantum algorithm for estimating multiple expectation values,

    W. J. Hugginset al., “Nearly optimal quantum algorithm for estimating multiple expectation values,”Physical Review Letters, 2022

    work page 2022

  37. [37]

    Rubini, P

    L. Rubiniet al., “Chronic Kidney Disease,” UCI Machine Learning Repository, 2015, DOI: https://doi.org/10.24432/C5G020

    work page doi:10.24432/c5g020 2015

  38. [38]

    Chronic kidney disease dataset,

    R. E. Kharoua, “Chronic kidney disease dataset,” https://www.kaggle.com/ datasets/rabieelkharoua/chronic-kidney-disease-dataset-analysis, 2024

    work page 2024

  39. [39]

    Risk Factor Prediction of Chronic Kidney Disease,

    M. A. Islam and S. Akter, “Risk Factor Prediction of Chronic Kidney Disease,” UCI Machine Learning Repository, 2020, DOI: https://doi.org/10.24432/C5WP64

    work page doi:10.24432/c5wp64 2020

  40. [40]

    PennyLane: Automatic differentiation of hybrid quantum-classical computations

    V . Bergholmet al., “Pennylane: Automatic differentiation of hybrid quantum-classical computations,”arXiv:1811.04968, 2022

    work page internal anchor Pith review arXiv 2022

  41. [41]

    The mcc-f1 curve: a performance evaluation technique for binary classification,

    C. Caoet al., “The mcc-f1 curve: a performance evaluation technique for binary classification,”arXiv:2006.11278, 2020

    work page arXiv 2006

  42. [42]

    General performance score for classification problems,

    I. M. de Diegoet al., “General performance score for classification problems,”Applied Intelligence, 2022

    work page 2022

  43. [43]

    Deep roc analysis and auc as balanced average accuracy, for improved classifier selection, audit and explanation,

    A. M. Carringtonet al., “Deep roc analysis and auc as balanced average accuracy, for improved classifier selection, audit and explanation,”TPAMI, 2023

    work page 2023