Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Automated screening over a machine-generated circuit library can select a small six-qubit quantum circuit that reaches 97% test accuracy and 0.9946 ROC-AUC on credit-card fraud, outperforming manually designed quantum models.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

CircuitHunt screens KetGPT circuits with qubit/parameter filters and 5-epoch macro-F1 scoring, selecting circuit #221 (6 qubits, 9 parameters) that reportedly hits 97% accuracy, but SMOTE-before-split inflates the test numbers.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection The screening pipeline idea is useful, but SMOTE-before-split makes the headline numbers invalid. the 3 major comments →

arxiv 2508.21366 v1 pith:ZYVRKXHR submitted 2025-08-29 quant-ph

CircuitHunt: Automated Quantum Circuit Screening for Superior Credit-Card Fraud Detection

classification quant-ph
keywords automated quantum circuit screeningquantum architecture searchhybrid quantum neural networksvariational quantum circuitscredit card fraud detectionmacro-F1 checkpointingSMOTEimbalanced classification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper argues that the bottleneck in quantum machine learning for fraud detection is not just data imbalance but the manual design of quantum circuits. It introduces CircuitHunt, an automated pipeline that filters a pre-generated library of 1,000 quantum circuits down to candidates respecting qubit and parameter budgets, embeds each in a standardized hybrid quantum-classical network, and ranks them by validation macro-F1 after short training runs. The top-ranked six-qubit circuit with nine trainable parameters is retrained and, on the credit-card fraud benchmark, is reported to reach 97% test accuracy, 0.97 macro-F1, and 0.9946 ROC-AUC, beating four hand-designed QML baselines. The stakes: if the result holds, automated screening could replace manual quantum architecture design in resource-constrained, real-world applications.

Core claim

CircuitHunt is presented as a solution to the architectural bottleneck in quantum machine learning: instead of hand-designing a variational circuit for a fraud-detection task, the pipeline treats a library of machine-generated circuits as a search space. It keeps only circuits with 3–10 qubits, at least one trainable gate, at most 30 trainable parameters, and a finite expectation value; embeds each survivor in one standardized hybrid quantum-classical network with a learnable residual connection; trains each briefly for five epochs and checkpoints on validation macro-F1; then retrains the winner for 20 epochs. The selected circuit, six qubits and nine trainable parameters, is reported to ach

What carries the argument

The central mechanism is the macro-F1 checkpointing loop wrapped around a fixed hybrid QNN template. Each filtered circuit is dropped into the same template—a classical encoder that maps 28 features to the circuit's qubit count, rotation-gate angle encoding, the candidate circuit followed by entangling CNOT chains, a learnable residual term z_res = z_quantum + α z_classical, and a post-quantum classifier—so differences in performance are attributed to the circuit rather than to ad hoc architecture choices. The five-epoch validation macro-F1 acts as a cheap proxy that lets the pipeline discard weak circuits early and keep only the best checkpoint.

Load-bearing premise

The load-bearing premise is that the SMOTE-balanced dataset split still yields a test set that represents real, previously unseen transactions; if synthetic minority samples generated before the split appear in validation and test, the reported 97% accuracy is partly a measure of interpolation near the training distribution.

What would settle it

Re-run the pipeline with SMOTE applied only to the training fold after the split, then evaluate on the untouched original test set; compare accuracy and macro-F1 to the reported 97% and ~0.97. A large drop would demonstrate that the headline numbers are inflated by synthetic test points near the training distribution rather than by real unseen transactions.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Architecture search for QML on imbalanced financial data becomes a screening problem rather than a design problem: same template, filtered library, short validation runs.
  • The learnable skip connection is load-bearing in the reported result: removing it drops accuracy from ~97% to ~70% and ROC-AUC from 0.9946 to 0.7706 in the paper's ablation.
  • Ranking on macro-F1 rather than accuracy aligns circuit selection with the minority class, which is the class that matters in fraud detection.
  • The selected six-qubit, nine-parameter architecture is a concrete, reproducible candidate for others working on imbalanced tabular classification.

Where Pith is reading between the lines

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

  • A natural re-run the paper does not include: apply SMOTE only inside the training fold, evaluate on untouched original test transactions, and compare to the reported 97%; the size of that gap would isolate how much of the result depends on synthetic leakage.
  • The pipeline's logic transfers to other imbalanced tabular tasks (network intrusion, medical screening) by swapping dataset and encoder, but the transfer is only plausible, not established, because the paper tests a single benchmark.
  • The paper leaves the five-epoch-to-20-epoch ranking assumption unvalidated; computing rank correlation between short and full training would show whether the screening proxy is safe or needs a two-stage schedule.
  • Since the skip connection dominates the ablation, a purely classical version of the residual network (same encoder, skip, and post-classifier, no quantum module) should be run as a sanity check to measure the quantum layer's marginal contribution.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper presents CircuitHunt, an automated workflow for selecting quantum circuits from the KetGPT dataset for a hybrid quantum-classical fraud-detection model. After filtering circuits by qubit and parameter constraints, each candidate is embedded in a standardized QNN with a learnable residual connection, trained for 5 epochs, and ranked by validation macro-F1. The top circuit (#221, 6 qubits, 9 parameters) is then trained for 20 epochs and reported to achieve 97% accuracy, macro-F1 above 0.965, and ROC-AUC 0.9946 on the Credit Card Fraud dataset, outperforming the QML baselines EstimatorQNN, QGNN, SQNN, and QFDNN. An ablation shows that removing the residual connection degrades performance, and the paper claims that the screening reduces design time from days to hours.

Significance. If the reported evaluation were sound, CircuitHunt would be a useful demonstration of automated, task-driven quantum architecture search for an imbalanced financial classification problem. The paper has concrete strengths: a well-specified screening algorithm (Algorithm 1), a fixed model template for fair candidate comparison, and a clear ablation study. However, the central empirical claim is invalidated by the preprocessing protocol: SMOTE before splitting contaminates the held-out sets and the selection objective, and the test distribution is artificially balanced. Because the quantum advantage and the specific circuit choice rest on these numbers, the significance of the contribution is not currently established. The underlying idea is salvageable with a correctly re-run evaluation, but the presented evidence does not support the headline.

major comments (3)
  1. [III-A, Algorithm 1 (lines 2–4), IV-B] Algorithm 1 (lines 2–4) and Section III-A apply SMOTE before the train/val/test split and then select equal numbers of samples per class. Two problems follow. First, SMOTE generates synthetic minority points from full-data neighborhoods; because the split occurs afterward, validation and test contain synthetic points whose nearest neighbors lie in the training set. Table II, Fig. 3, and the selection objective Eq. (6) are therefore contaminated, so the reported 97% accuracy and macro-F1 are not estimates for unseen real transactions. Second, the test set is artificially balanced (50/50) while the original fraud rate is ~0.17%; accuracy on this test distribution does not describe real operating conditions. Section V lists limitations but omits both issues. The pipeline must be re-run with SMOTE and the MinMaxScaler fit on the training split only, and with the test set preserving the origi
  2. [IV-D, Table IV] The baselines EstimatorQNN, QGNN, SQNN, and QFDNN are compared without any statement of whether they were re-implemented under identical data handling, splits, training budget, and metric computation. Quoting numbers from previous papers is not controlled, and the 'outperforms all' claim is load-bearing. A fair comparison must run all baselines on the corrected pipeline. In addition, no classical baseline is reported; for the title's 'superior fraud detection' claim, a classical reference (e.g., XGBoost or logistic regression) is needed to establish practical relevance.
  3. [III-C, Eq. (6)] The selection criterion is validation macro-F1 after 5 epochs, while final performance is reported after 20 epochs. No rank-correlation or stability analysis is given to show that the short-run ranking predicts the long-run ranking. Without this, the choice of circuit #221 may be an artifact of early-training noise, and the claimed screening efficiency is unvalidated. The authors should report the correlation between 5-epoch and 20-epoch validation scores, or select at full training.
minor comments (5)
  1. [III-B, Eq. (3)] The condition |f_i(θ)| ≤ 1 is automatically satisfied for a Pauli-Z expectation value on any valid quantum state; it does not certify trainability. Either remove this condition or rephrase it as a simulator sanity check.
  2. [III-A] The sentence 'an equal number of samples are then randomly selected from each class' should specify whether this selection occurs before or after SMOTE and report the resulting dataset sizes. As written, the imbalance-handling procedure is ambiguous.
  3. [IV-A, Table I] The runtime claim 'from days to hours' in the abstract is not quantified anywhere. Report actual wall-clock time for the screening stage (e.g., per-circuit training time, total search time) to support the scalability claim.
  4. [Figures 3 and 4] The captions use informal markers ('①', '②') and phrases such as 'Random classifier baseline (FPR=TPR)' without fully labeled axes and legends. The figures should be self-contained for a reader.
  5. [References] References [42] and [44] both point to the KetGPT resource; use a single canonical citation. Also add a code/data availability statement, since the text claims a 'reproducible workflow' but no repository is listed.

Circularity Check

1 steps flagged

Reported test performance is not independent of the selection criterion: SMOTE is applied before the train/val/test split, so the 'held-out' test set contains synthetic points generated from training-region neighborhoods; the headline macro-F1 is therefore partially a re-measurement of the model-selection objective.

specific steps
  1. fitted input called prediction [Section III-A (Data Preprocessing) and Algorithm 1, lines 2-4; selection in Eq. (6); test evaluation in Section IV-B/Table II]
    "we apply SMOTE to increase the representation of the minority class ... The normalized dataset is subsequently partitioned into training, validation, and test sets using stratified sampling ... Algorithm 1: 2 Apply SMOTE to balance the dataset ... 4 Split into training, validation, and test sets: {Xtrain, Xval, Xtest, ytrain, yval, ytest}."

    SMOTE creates synthetic minority-class vectors by interpolating among nearest neighbors. Because it is applied before the split, the validation and test partitions inherit synthetic points whose nearest neighbors lie in the training partition. The selection rule (Eq. 6) chooses C* = arg max F1_val, and the paper then reports the macro-F1 on that same contaminated test set (Table II). The test macro-F1 therefore is not an estimate on real unseen transactions; it partly re-evaluates the very objective used for selection on synthetic data drawn from training-region neighborhoods. The 97% 'prediction' is inflated by the construction of the preprocessing order, not demonstrated by the circuit's generalization. This is a fitted input (SMOTE-augmented data) being called a held-out prediction.

full rationale

The screening pipeline itself is not circular: candidates are filtered by qubit/parameter constraints (Eqs. 1-3), and circuit #221 is selected by validation macro-F1 (Eq. 6), which is a legitimate selection criterion if validation data are independent. The model architecture and ablation are self-contained. However, the evaluation protocol violates that independence: SMOTE before the split contaminates both validation and test sets with synthetic interpolations of training data. Consequently the headline test macro-F1/accuracy is not an honest held-out measurement; it partially re-measures the selection objective. The self-citations to earlier QML fraud models are used as comparison baselines, not as load-bearing uniqueness/ansatz justifications, so they do not add circularity beyond the evaluation leakage. Score 6 reflects a partial but central circularity: the reported prediction is not independent of the constructed input.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central claim rests on a set of hand-chosen budget thresholds and training hyperparameters (none of them ablated), on the imported validity of the KetGPT circuit library, on the assumption that 5-epoch macro-F1 ranking predicts final performance, and on the processing decision to apply SMOTE before the data split. No new physical entities are postulated; the search is over prior-work circuits, so the invented-entity burden is zero.

free parameters (4)
  • Screening budget thresholds = qubit window [3,10]; P_max=30; 5-epoch screening runs
    Hand-chosen in Section III-B and Table I; no sensitivity analysis showing circuit rankings are stable across these thresholds.
  • SMOTE rebalancing target = 10,000 samples per class
    Section III-A and Table I; arbitrary target that defines the evaluation distribution and inflates minority-class representation.
  • Residual connection initial weight = alpha = 0.1
    Table I; chosen by hand, no justification or ablation on the initialization.
  • Fixed training hyperparameters = LR 0.01; batch 32; pre-NN width 64; post-NN width 16
    Table I; held fixed across candidates and never ablated, yet these choices shape the macro-F1 ranking.
axioms (4)
  • domain assumption A 5-epoch macro-F1 ranking at LR 0.01 reliably predicts the ranking after 20-epoch full training
    Section III-C and Fig. 1: the entire selection procedure assumes short-training validation macro-F1 is a faithful proxy for final test performance; no rank-correlation evidence is given.
  • domain assumption The KetGPT dataset is a valid search space of quantum circuits
    Section III-B imports the 1,000 transformer-generated circuits from [19] as 'structurally valid quantum circuits' without independent validation for this task.
  • ad hoc to paper Zero-initialization finiteness (Eq. 3, |f_i(theta)| <= 1) certifies executability and trainability
    Section III-B; the check excludes only grossly invalid circuits and says nothing about gradient behavior or trainability.
  • domain assumption RX angle encoding over [0, pi] captures the information in the 28 PCA features
    Section III-A and Table I; standard but unvalidated on this benchmark, and the encoding pipeline contributes to the reported performance.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of CircuitHunt: Automated Quantum Circuit Screening for Superior Credit-Card Fraud Detection." pith.science (2026). https://pith.science/paper/ZYVRKXHR

@misc{pith2026250821366,
  author       = {Pith},
  title        = {Pith review of: CircuitHunt: Automated Quantum Circuit Screening for Superior Credit-Card Fraud Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZYVRKXHR}},
  note         = {Machine review of arXiv:2508.21366}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Designing effective quantum models for real-world tasks remains a key challenge within Quantum Machine Learning (QML), particularly in applications such as credit card fraud detection, where extreme class imbalance and evolving attack patterns demand both accuracy and adaptability. Most existing approaches rely on either manually designed or randomly initialized circuits, leading to high failure rates and limited scalability. In this work, we introduce CircuitHunt, a fully automated quantum circuit screening framework that streamlines the discovery of high-performing models. CircuitHunt filters circuits from the KetGPT dataset using qubit and parameter constraints, embeds each candidate into a standardized hybrid QNN, and performs rapid training with checkpointing based on macro-F1 scores to discard weak performers early. The top-ranked circuit is then fully trained, achieving 97% test accuracy and a high macro-F1 score on a challenging fraud detection benchmark. By combining budget-aware pruning, empirical evaluation, and end-to-end automation, CircuitHunt reduces architecture search time from days to hours while maintaining performance. It thus provides a scalable and task-driven tool for QML deployment in critical financial applications.

Figures

Figures reproduced from arXiv: 2508.21366 by Akshat Singh, Muhammad Shafique, Nouhaila Innan.

Figure 1
Figure 1. Figure 1: End-to-end pipeline for hybrid quantum-classical circuit evaluation. (1) The credit card fraud dataset is preprocessed with SMOTE-based balancing, scaling, and split into training, validation, and test sets. (2) Candidate quantum circuits are filtered from the KetGPT dataset by qubit count (3–10), presence of trainable gates, and parameter budget (≤ 30). (3) Each candidate circuit is evaluated using a hybr… view at source ↗
Figure 3
Figure 3. Figure 3: Performance evaluation of the CircuitHunt-selected model. (a) Accuracy trends over 20 epochs show early stabilization of validation accuracy and a narrowing gap with training accuracy, indicating effective generalization. (b) Loss curves exhibit a sharp initial decline in validation loss, followed by smooth convergence, reflecting stable and well-tuned optimization. (c) Macro-F1 score results demonstrate c… view at source ↗
Figure 4
Figure 4. Figure 4: Performance evaluation of the ablated model without the residual skip connection. (a) Training and validation accuracy curves indicate slower learning and a persistent generalization gap, underscoring the importance of the skip connection for stable convergence. (b) Loss curves show steady but suboptimal convergence, with a consistently higher validation loss compared to training. (c) Macro-F1 score trends… view at source ↗

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. Graph-Based Bayesian Optimization for Quantum Circuit Architecture Search with Uncertainty Calibrated Surrogates

    quant-ph 2025-12 conditional novelty 5.0

    A graph-neural-network-guided Bayesian search discovers compact variational quantum circuits for a cybersecurity classification task, slightly beating a flat-feature MLP surrogate and dominating fixed-architecture baselines.

Reference graph

Works this paper leans on

46 extracted references · 29 canonical work pages · cited by 1 Pith paper · 3 internal anchors

  1. [1]

    A systematic review of ai-enhanced techniques in credit card fraud detection,

    I. Y . Hafez, A. Y . Hafez, A. Saleh, A. A. Abd El-Mageed, and A. A. Abohany, “A systematic review of ai-enhanced techniques in credit card fraud detection,” Journal of Big Data , vol. 12, no. 1, p. 6, 2025

  2. [2]

    Anomaly detection in online credit card data using optimized multi-view heterogeneous graph neural networks,

    T. J. Berkmans and S. Karthick, “Anomaly detection in online credit card data using optimized multi-view heterogeneous graph neural networks,” Knowledge-Based Systems, p. 113767, 2025

  3. [3]

    Mixed quantum–classical method for fraud detection with quantum feature selection,

    M. Grossi, N. Ibrahim, V . Radescu, R. Loredo, K. V oigt, C. V on Altrock, and A. Rudnik, “Mixed quantum–classical method for fraud detection with quantum feature selection,” IEEE Transactions on Quantum Engi- neering, vol. 3, pp. 1–12, 2022

  4. [4]

    Evaluating the computational advantages of the variational quantum circuit model in financial fraud detection,

    A. Tudisco, D. V olpe, G. Ranieri, G. Curato, D. Ricossa, M. Graziano, and D. Corbelletto, “Evaluating the computational advantages of the variational quantum circuit model in financial fraud detection,” IEEE Access, 2024

  5. [5]

    Smote: synthetic minority over-sampling technique,

    N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: synthetic minority over-sampling technique,” Journal of Artificial Intelligence Research, vol. 16, pp. 321–357, 2002

  6. [6]

    Quantum machine learning,

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, “Quantum machine learning,” Nature, vol. 549, no. 7671, pp. 195–202, 2017

  7. [7]

    Better than classical? the subtle art of benchmarking quantum machine learning models,

    J. Bowles, S. Ahmed, and M. Schuld, “Better than classical? the subtle art of benchmarking quantum machine learning models,” arXiv preprint arXiv:2403.07059, 2024

  8. [8]

    The power of quantum neural networks,

    A. Abbas, D. Sutter, C. Zoufal, A. Lucchi, A. Figalli, and S. Woerner, “The power of quantum neural networks,” Nature Computational Science, vol. 1, no. 6, pp. 403–409, 2021

  9. [9]

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

    N. Innan, M. Kashif, A. Marchisio, M. Bennai, and M. Shafique, “Next- generation quantum neural networks: Enhancing efficiency, security, and privacy,” in 2025 IEEE 31st International Symposium on On-Line Testing and Robust System Design (IOLTS) . IEEE, 2025, pp. 1–4

  10. [10]

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

    M. Kashif, A. Marchisio, and M. Shafique, “Computational advantage in hybrid quantum neural networks: Myth or reality?” arXiv preprint arXiv:2412.04991, 2024

  11. [11]

    Quantum neural networks: A comparative analysis and noise robustness evaluation,

    T. Ahmed, M. Kashif, A. Marchisio, and M. Shafique, “Quantum neural networks: A comparative analysis and noise robustness evaluation,” arXiv preprint arXiv:2501.14412, 2025

  12. [12]

    Quantum convo- lutional neural network based on variational quantum circuits,

    L.-H. Gong, J.-J. Pei, T.-F. Zhang, and N.-R. Zhou, “Quantum convo- lutional neural network based on variational quantum circuits,” Optics Communications, vol. 550, p. 129993, 2024

  13. [13]

    Enhancing quantum support vector machines through variational kernel training,

    N. Innan, M. A.-Z. Khan, B. Panda, and M. Bennai, “Enhancing quantum support vector machines through variational kernel training,” Quantum Information Processing, vol. 22, no. 10, p. 374, 2023

  14. [14]

    More: Measurement and correlation based variational quantum circuit for multi-classification,

    J. Wu, T. Hu, and Q. Li, “More: Measurement and correlation based variational quantum circuit for multi-classification,” in 2023 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 1. IEEE, 2023, pp. 208–218

  15. [15]

    A variational quantum perceptron with grover’s algorithm for efficient classification,

    N. Innan and M. Bennai, “A variational quantum perceptron with grover’s algorithm for efficient classification,” Physica Scripta, vol. 99, no. 5, p. 055120, 2024

  16. [16]

    Lep-qnn: Loan eligibility prediction using quantum neural networks,

    N. Innan, A. Marchisio, M. Bennai, and M. Shafique, “Lep-qnn: Loan eligibility prediction using quantum neural networks,” arXiv preprint arXiv:2412.03158, 2024

  17. [17]

    Quantum bayesian networks for machine learning in oil-spill detection,

    O. I. Siddiqui, N. Innan, A. Marchisio, M. Bennai, and M. Shafique, “Quantum bayesian networks for machine learning in oil-spill detection,” arXiv preprint arXiv:2412.19843 , 2024

  18. [18]

    Sentiqnf: A novel approach to sentiment analysis using quantum algorithms and neuro-fuzzy systems,

    K. Dave, N. Innan, B. K. Behera, Z. Mumtaz, S. Al-Kuwari, and A. Farouk, “Sentiqnf: A novel approach to sentiment analysis using quantum algorithms and neuro-fuzzy systems,” IEEE Transactions on Computational Social Systems , 2025

  19. [19]

    Ketgpt–dataset augmentation of quantum circuits using transformers,

    B. Apak, M. Bandic, A. Sarkar, and S. Feld, “Ketgpt–dataset augmentation of quantum circuits using transformers,” arXiv preprint arXiv:2402.13352, 2024

  20. [20]

    Quantum state tomography using quantum machine learning,

    N. Innan, O. I. Siddiqui, S. Arora, T. Ghosh, Y . P. Ko c ¸ak, D. Paragas, A. A. O. Galib, M. A.-Z. Khan, and M. Bennai, “Quantum state tomography using quantum machine learning,” Quantum Machine Intelligence, vol. 6, no. 1, p. 28, 2024

  21. [21]

    Quantum clustering for cybersecurity,

    W. El Maouaki, N. Innan, A. Marchisio, T. Said, M. Bennai, and M. Shafique, “Quantum clustering for cybersecurity,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 2. IEEE, 2024, pp. 5–10

  22. [22]

    Optimizing low-energy carbon iiot systems with quantum algorithms: Performance evaluation and noise robustness,

    K. Dave, N. Innan, B. K. Behera, S. Mumtaz, S. Al-Kuwari, A. Farouk et al. , “Optimizing low-energy carbon iiot systems with quantum algorithms: Performance evaluation and noise robustness,” IEEE Internet of Things Journal , 2025

  23. [23]

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

    N. Innan, B. K. Behera, S. Al-Kuwari, and A. Farouk, “Qnn-vrcs: A quantum neural network for vehicle road cooperation systems,” IEEE Transactions on Intelligent Transportation Systems , 2025

  24. [24]

    Resource allocation optimization in 5g networks using variational quantum regressor,

    P. Pathak, V . Oad, A. Prajapati, and N. Innan, “Resource allocation optimization in 5g networks using variational quantum regressor,” in 2024 International Conference on Quantum Communications, Networking, and Computing (QCNC) . IEEE, 2024, pp. 101–105

  25. [25]

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

    S. Dutta, N. Innan, K. Najafi, S. B. Yahia, and M. Shafique, “Quiet- sr: Quantum image enhancement transformer for single image super- resolution,” arXiv preprint arXiv:2503.08759 , 2025

  26. [26]

    Qadqn: Quantum attention deep q-network for financial market prediction,

    S. Dutta, N. Innan, A. Marchisio, S. B. Yahia, and M. Shafique, “Qadqn: Quantum attention deep q-network for financial market prediction,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 2. IEEE, 2024, pp. 341–346

  27. [27]

    Hqnn-fsp: A hybrid classical-quantum neural network for regression-based financial stock market prediction,

    P. K. Choudhary, N. Innan, M. Shafique, and R. Singh, “Hqnn-fsp: A hybrid classical-quantum neural network for regression-based financial stock market prediction,” arXiv preprint arXiv:2503.15403 , 2025

  28. [28]

    A brief review of quantum machine learning for financial services,

    M. Doosti, P. Wallden, C. B. Hamill, R. Hankache, O. T. Brown, and C. Heunen, “A brief review of quantum machine learning for financial services,” arXiv preprint arXiv:2407.12618 , 2024

  29. [29]

    Financial fraud detection using quantum graph neural networks,

    N. Innan, A. Sawaika, A. Dhor, S. Dutta, S. Thota, H. Gokal, N. Patel, M. A.-Z. Khan, I. Theodonis, and M. Bennai, “Financial fraud detection using quantum graph neural networks,” Quantum Machine Intelligence , vol. 6, no. 1, Feb. 2024. [Online]. Available: http://dx.doi.org/10.1007/s42484-024-00143-6

  30. [30]

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

    N. Innan, M. A.-Z. Khan, and M. Bennai, “Financial fraud detection: a comparative study of quantum machine learning models,” International Journal of Quantum Information , vol. 22, no. 02, p. 2350044, 2024

  31. [31]

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

    M. Alami, N. Innan, M. Shafique, and M. Bennai, “Comparative performance analysis of quantum machine learning architectures for credit card fraud detection,” dec 2024. [Online]. Available: https://arxiv.org/abs/2412.19441

  32. [32]

    Toward practical quantum machine learning: A novel hybrid quantum lstm for fraud detection,

    R. Ubale, S. Deshpande, G. T. Byrd et al., “Toward practical quantum machine learning: A novel hybrid quantum lstm for fraud detection,” arXiv preprint arXiv:2505.00137 , 2025

  33. [33]

    Qfnn-ffd: Quantum federated neural network for financial fraud detection,

    N. Innan, A. Marchisio, M. Bennai, and M. Shafique, “Qfnn-ffd: Quantum federated neural network for financial fraud detection,” arXiv preprint arXiv:2404.02595, 2024

  34. [34]

    A privacy-preserving federated framework with hybrid quantum-enhanced learning for financial fraud detection,

    A. Sawaika, S. Krishna, T. Tomar, D. P. Suggisetti, A. Lal, T. Shrivastav, N. Innan, and M. Shafique, “A privacy-preserving federated framework with hybrid quantum-enhanced learning for financial fraud detection,” arXiv preprint arXiv:2507.22908 , 2025

  35. [35]

    Differentiable quantum architecture search,

    S.-X. Zhang, C.-Y . Hsieh, S. Zhang, and H. Yao, “Differentiable quantum architecture search,” Quantum Science and Technology , vol. 7, no. 4, p. 045023, 2022

  36. [36]

    Qas-qtns: Curriculum reinforcement learning-driven quantum architecture search for quantum tensor networks,

    S. Dutta, N. Innan, S. B. Yahia, and M. Shafique, “Qas-qtns: Curriculum reinforcement learning-driven quantum architecture search for quantum tensor networks,” arXiv preprint arXiv:2507.12013 , 2025

  37. [37]

    Quantum circuit architecture search for variational quantum algorithms,

    Y . Du, T. Huang, S. You, M.-H. Hsieh, and D. Tao, “Quantum circuit architecture search for variational quantum algorithms,” npj Quantum Information, vol. 8, no. 1, p. 62, 2022

  38. [38]

    Quantum Architecture Search: A Survey

    D. Martyniuk, J. Jung, and A. Paschke, “Quantum architecture search: A survey,” 2024. [Online]. Available: https://arxiv.org/abs/2406.06210

  39. [39]

    Quantum Architecture Search via Continual Reinforcement Learning

    E. Ye and S. Y .-C. Chen, “Quantum architecture search via continual reinforcement learning,” arXiv preprint arXiv:2112.05779 , 2021

  40. [40]

    Quantum architecture search via deep reinforcement learning,

    E.-J. Kuo, Y .-L. L. Fang, and S. Y .-C. Chen, “Quantum architecture search via deep reinforcement learning,” arXiv preprint arXiv:2104.07715, 2021

  41. [41]

    Challenges for reinforcement learning in quantum circuit design,

    P. Altmann, J. Stein, M. K ¨olle, A. B ¨arligea, M. Zorn, T. Gabor, T. Phan, S. Feld, and C. Linnhoff-Popien, “Challenges for reinforcement learning in quantum circuit design,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE) , vol. 1. IEEE, 2024, pp. 1600–1610

  42. [42]

    Ketgpt quantum circuit dataset,

    PennyLane AI, “Ketgpt quantum circuit dataset,” https://pennylane.ai/ datasets/ketgpt, 2024, accessed 2024-05-11

  43. [43]

    Pennylane: Automatic differentiation of hybrid quantum-classical computations,

    V . Bergholm, J. Izaac, M. Schuld, C. Gogolin, S. Ahmed, V . Ajith, M. S. Alam, G. Alonso-Linaje, B. AkashNarayanan, A. Asadi et al., “Pennylane: Automatic differentiation of hybrid quantum-classical computations,” arXiv preprint arXiv:1811.04968 , 2018

  44. [44]

    Ketgpt data

    “Ketgpt data.” [Online]. Available: https://www.kaggle.com/datasets/ boranapak/ketgpt-data

  45. [45]

    Credit card fraud detection

    “Credit card fraud detection.” [Online]. Available: https://www.kaggle. com/datasets/mlg-ulb/creditcardfraud/data

  46. [46]

    QFDNN: A Resource-Efficient Variational Quantum Feature Deep Neural Networks for Fraud Detection and Loan Prediction

    S. Das, A. Meghanath, B. K. Behera, S. Mumtaz, S. Al-Kuwari, and A. Farouk, “Qfdnn: A resource-efficient variational quantum feature deep neural networks for fraud detection and loan prediction,” arXiv preprint arXiv:2504.19632, 2025

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.