pith. sign in

arxiv: 2604.15048 · v1 · submitted 2026-04-16 · 🪐 quant-ph

GAT-QNN: Genetic Algorithm-Based Training of Hybrid Quantum Neural Networks

Tasnim Ahmed , Alberto Marchisio , Muhammad Kashif , Nouhaila Innan , Muhammad Shafique This is my paper

Pith reviewed 2026-05-10 11:29 UTC · model grok-4.3

classification 🪐 quant-ph
keywords hybrid quantum neural networksgenetic algorithmquantum circuit architectureNISQ devicesparameterized quantum circuitsMNIST classificationmicroCircuit selection
0
0 comments X

The pith

Genetic algorithm training lets hybrid quantum networks select optimal subcircuits for any backend without retraining each candidate

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

The paper establishes that hybrid quantum neural networks can be trained more effectively by using a genetic algorithm to explore circuit architectures through a macroCircuit that contains all possible subcircuits. By sampling microCircuits, training them, and returning their parameters to the macroCircuit, the framework builds a reusable set of weights. A subsequent inference stage then applies another genetic algorithm to choose the best microCircuits for any given quantum backend using those weights. This matters because it sidesteps the need to retrain separate models for every possible circuit and backend combination, which would otherwise be computationally prohibitive. The method also supports deploying compact circuits to lower resource demands while reporting accuracy improvements of 22 to 23 percent on a four-class version of the MNIST dataset.

Core claim

GAT-QNN trains hybrid quantum neural networks by iteratively sampling microCircuits from a macroCircuit, training the samples, and reintegrating the learned parameters. After this training phase, a separate genetic algorithm uses the fixed macroCircuit weights to score and select promising microCircuits for deployment on chosen backends, achieving backend-aware optimization without retraining candidates.

What carries the argument

The macroCircuit as a shared search space and parameter reservoir from which microCircuits are sampled during training and evaluated during inference.

If this is right

  • The two-stage process supports backend-aware microCircuit selection without retraining each candidate architecture.
  • Smaller microCircuits derived from the macroCircuit can be deployed to reduce gate count and computational resources.
  • The approach delivers consistent 22-23 percent test accuracy gains for GA-driven inference across multiple backends on four-class MNIST classification.

Where Pith is reading between the lines

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

  • The parameter reintegration step may simplify adaptation when hardware noise profiles change between training and deployment.
  • The same macroCircuit could support rapid testing of circuit variants on new datasets without repeating the full training loop.
  • Success here suggests the method could scale to other variational quantum algorithms that face large discrete architecture spaces.

Load-bearing premise

Parameters learned from sampled microCircuits transfer effectively when reintegrated into the macroCircuit, allowing accurate performance evaluation of other microCircuits without retraining them individually.

What would settle it

If evaluating unsampled microCircuits with the reintegrated macroCircuit parameters produces accuracy no better than random selection or requires individual retraining to reach the reported levels on the four-class MNIST task.

Figures

Figures reproduced from arXiv: 2604.15048 by Alberto Marchisio, Muhammad Kashif, Muhammad Shafique, Nouhaila Innan, Tasnim Ahmed.

Figure 1
Figure 1. Figure 1: Hybrid quantum neural network (HQNN) pipeline used in this work: [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Genetic algorithm workflow adopted in this work. An initial population [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed GAT-QNN methodology. The macroCircuit defines a discrete architectural search space over layered RX (parameterized [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cross-backend test accuracy comparing GA-trained versus regularly [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
read the original abstract

Hybrid Quantum Neural Networks (HQNNs) combine classical learning with parameterized quantum circuits, but their practical performance is often limited by (i) the noise of Noisy Intermediate-Scale Quantum (NISQ) devices and (ii) the large, discrete design space of quantum circuit architectures. Moreover, HQNNs are commonly trained using a fixed circuit and a single backend, even though deployment frequently targets heterogeneous backends where compilation and execution characteristics may differ. To address these challenges, we propose GAT-QNN, a genetic algorithm (GA)-based framework that trains a macroCircuit (search space) by iteratively sampling microCircuits (subcircuits), training them, and reintegrating their learned parameters into the macroCircuit. After training, we run an independent GA-driven inference stage that evaluates candidate microCircuits using the trained macroCircuit weights and selects top-performing architectures for deployment. This two-stage approach enables backend-aware microCircuit selection without retraining each candidate architecture and can also reduce computational resources (gate count) by deploying smaller microCircuits derived from the macroCircuit. We validate the approach on MNIST classification (four classes) and report consistent 22-23% test accuracy gains for GA-driven inference across multiple backends.

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

2 major / 1 minor

Summary. The manuscript introduces GAT-QNN, a two-stage genetic algorithm framework for hybrid quantum neural networks. A macroCircuit (search space) is trained by iteratively sampling microCircuits (subcircuits), training each on the task, and reintegrating the learned parameters back into the macroCircuit. An independent GA-driven inference stage then uses the resulting macroCircuit weights to evaluate candidate microCircuits and select top-performing architectures for deployment on heterogeneous backends, with the additional benefit of deploying smaller derived microCircuits to reduce gate count. The approach is validated on four-class MNIST classification, reporting consistent 22-23% test accuracy gains for the GA-driven inference across multiple backends.

Significance. If the reintegration step produces weights that faithfully preserve relative performance rankings across microCircuit topologies, the method would provide a practical route to backend-aware architecture search and resource reduction in NISQ-era quantum machine learning without exhaustive per-architecture retraining. The separation of a training phase from an inference-time selection phase addresses real deployment constraints and could lower computational costs if the empirical claims are substantiated.

major comments (2)
  1. [Method and validation sections] The central claim that reintegrated parameters from sampled microCircuits serve as a reliable proxy for evaluating and ranking unseen microCircuits (abstract and method description) lacks supporting evidence. Different microCircuits possess distinct gate sets, topologies, and noise sensitivities; parameters optimized on one subset need not yield accurate relative loss estimates when applied to another, particularly under NISQ compilation and hardware variation. No ablation studies, ranking-correlation metrics, or theoretical argument are supplied to show that the proxy preserves ordering fidelity.
  2. [Abstract and experimental results] The reported 22-23% test accuracy gains (abstract) are presented without baselines, statistical significance tests, error bars, number of independent runs, or explicit definition of the reference (e.g., fixed-architecture HQNN, random microCircuit selection, or standard gradient training). Without these details the magnitude and robustness of the improvement cannot be assessed, weakening the empirical support for the two-stage procedure.
minor comments (1)
  1. Notation for macroCircuit versus microCircuit could be clarified with a diagram or explicit definition early in the text to aid readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and will make the indicated revisions to strengthen the empirical support and clarity of the work.

read point-by-point responses

  1. Referee: [Method and validation sections] The central claim that reintegrated parameters from sampled microCircuits serve as a reliable proxy for evaluating and ranking unseen microCircuits (abstract and method description) lacks supporting evidence. Different microCircuits possess distinct gate sets, topologies, and noise sensitivities; parameters optimized on one subset need not yield accurate relative loss estimates when applied to another, particularly under NISQ compilation and hardware variation. No ablation studies, ranking-correlation metrics, or theoretical argument are supplied to show that the proxy preserves ordering fidelity.

    Authors: We agree that direct validation of the proxy assumption is required. The original submission demonstrates consistent gains across backends but does not include explicit ablations or correlation metrics. In revision we will add a dedicated ablation subsection that evaluates ordering fidelity by comparing proxy-based rankings (using reintegrated macroCircuit weights) against rankings obtained from independent full training of each candidate microCircuit. We will report Spearman rank correlation coefficients computed over a held-out set of 50 microCircuits, together with the fraction of top-k selections that remain stable under the proxy. revision: yes

  2. Referee: [Abstract and experimental results] The reported 22-23% test accuracy gains (abstract) are presented without baselines, statistical significance tests, error bars, number of independent runs, or explicit definition of the reference (e.g., fixed-architecture HQNN, random microCircuit selection, or standard gradient training). Without these details the magnitude and robustness of the improvement cannot be assessed, weakening the empirical support for the two-stage procedure.

    Authors: We will revise both the abstract and the experimental results section to define the reference baseline explicitly as a fixed-architecture HQNN trained with standard gradient descent. We will report the number of independent runs (10), include error bars as standard deviation, and add paired t-test p-values to establish statistical significance. An additional baseline of random microCircuit selection will also be included for context. revision: yes

Circularity Check

0 steps flagged

No circularity: procedural GA framework with external empirical validation

full rationale

The paper presents GAT-QNN as a two-stage algorithmic procedure: iterative sampling of microCircuits from a macroCircuit, training those subcircuits, parameter reintegration, followed by an independent GA-driven inference stage for architecture selection. This is a search-and-optimization method, not a mathematical derivation chain. No equations, uniqueness theorems, or first-principles results are claimed that reduce by construction to fitted inputs or self-citations. The reported 22-23% accuracy gains on MNIST are presented as experimental outcomes across backends, not as predictions forced by the method's own definitions. The central assumption (that reintegrated parameters serve as a proxy for unseen microCircuits) is an empirical hypothesis subject to validation, not a tautological self-definition. The derivation chain is self-contained against external benchmarks and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based solely on the abstract, the central claim rests on standard assumptions of hybrid quantum-classical optimization and genetic search; no free parameters, invented entities, or non-standard axioms are explicitly introduced in the provided text.

axioms (1)
  • domain assumption Parameterized quantum circuits can be trained by classical optimization of their parameters
    Core premise of all hybrid quantum neural networks referenced in the abstract.

pith-pipeline@v0.9.0 · 5527 in / 1261 out tokens · 54969 ms · 2026-05-10T11:29:07.842351+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages

  1. [1]

    An introduction to quantum machine learning,

    M. Schuld, I. Sinayskiy, and F. Petruccione, “An introduction to quantum machine learning,”Contemporary Physics, 2015

    work page 2015

  2. [2]

    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?,” inDAC, 2025

    work page 2025

  3. [3]

    Po-qa: A framework for portfolio optimization using quantum algorithms,

    K. Zamanet al., “Po-qa: A framework for portfolio optimization using quantum algorithms,” inQCE, 2024

    work page 2024

  4. [4]

    Quav: Quantum-assisted path planning and optimization for uav navigation with obstacle avoidance,

    N. Innanet al., “Quav: Quantum-assisted path planning and optimization for uav navigation with obstacle avoidance,” inQAI, 2025

    work page 2025

  5. [5]

    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

  6. [6]

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

    S. Duttaet al., “Qadqn: Quantum attention deep q-network for financial market prediction,” inQCE, 2024

    work page 2024

  7. [7]

    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,” inQCE, 2025

    work page 2025

  8. [8]

    Quantum vs. classical machine learning: A benchmark study for financial prediction,

    R. Ahmadet al., “Quantum vs. classical machine learning: A benchmark study for financial prediction,”arXiv:2601.03802, 2026

    work page arXiv 2026

  9. [9]

    Quantum state tomography using quantum machine learning,

    N. Innanet al., “Quantum state tomography using quantum machine learning,”Quantum Machine Intelligence, vol. 6, no. 1, p. 28, 2024

    work page 2024

  10. [10]

    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

  11. [11]

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

    N. Innanet al., “Qfnn-ffd: Quantum federated neural network for financial fraud detection,” inQSW, 2025

    work page 2025

  12. [12]

    Fl-qdsnns: Federated learning with quantum dynamic spiking neural networks,

    N. Innan, A. Marchisio, and M. Shafique, “Fl-qdsnns: Federated learning with quantum dynamic spiking neural networks,” inQAI, 2025

    work page 2025

  13. [13]

    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

  14. [14]

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

    O. I. Siddiquiet al., “Quantum bayesian networks for machine learning in oil-spill detection,” inIJCNN, 2025

    work page 2025

  15. [15]

    Robqfl: Robust quantum federated learning in adversarial environment,

    W. El Maouakiet al., “Robqfl: Robust quantum federated learning in adversarial environment,” inQAI, 2025

    work page 2025

  16. [16]

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

    K. Daveet al., “Sentiqnf: A novel approach to sentiment analysis using quantum algorithms and neuro-fuzzy systems,”IEEE Transactions on Computational Social Systems, 2025

    work page 2025

  17. [17]

    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 preprint arXiv:2503.08759, 2025

    work page arXiv 2025

  18. [18]

    Design space exploration of hybrid quantum neural networks for chronic kidney disease,

    M. Kashifet al., “Design space exploration of hybrid quantum neural networks for chronic kidney disease,” inIJCNN, 2026

    work page 2026

  19. [19]

    Parameterized quantum circuits as machine learning models,

    M. Benedettiet al., “Parameterized quantum circuits as machine learning models,”Quantum science and technology, 2019

    work page 2019

  20. [20]

    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

  21. [21]

    Algorithms for quantum computation: discrete logarithms and factoring,

    P. W. Shor, “Algorithms for quantum computation: discrete logarithms and factoring,” inFOCS, 1994

    work page 1994

  22. [22]

    A fast quantum mechanical algorithm for database search,

    L. K. Grover, “A fast quantum mechanical algorithm for database search,” inSTOC, 1996

    work page 1996

  23. [23]

    Investigating the effect of noise on the training performance of hybrid quantum neural networks,

    M. Kashif, E. Sychiuco, and M. Shafique, “Investigating the effect of noise on the training performance of hybrid quantum neural networks,” inIJCNN, 2024

    work page 2024

  24. [24]

    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

  25. [25]

    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

  26. [26]

    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

  27. [27]

    Quantum computing in the nisq era and beyond,

    J. Preskill, “Quantum computing in the nisq era and beyond,”Quantum, 2018

    work page 2018

  28. [28]

    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

  29. [29]

    Quantum circuit learning,

    K. Mitaraiet al., “Quantum circuit learning,”Physical Review A, 2018

    work page 2018

  30. [30]

    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

  31. [31]

    Algorithm-oriented qubit mapping for variational quantum algorithms,

    Y . Jiet al., “Algorithm-oriented qubit mapping for variational quantum algorithms,”Physical Review Applied, 2025

    work page 2025

  32. [32]

    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

  33. [33]

    Quantumnas: Noise-adaptive search for robust quantum circuits,

    H. Wanget al., “Quantumnas: Noise-adaptive search for robust quantum circuits,” inHPCA, 2022

    work page 2022

  34. [34]

    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

  35. [35]

    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,” inIOLTS, 2025

    work page 2025

  36. [36]

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

    K. Zamanet al., “Studying the impact of quantum-specific hyperparameters on hybrid quantum-classical neural networks,” inWorld Congress in Computer Science, Computer Engineering & Applied Computing, 2024

    work page 2024

  37. [37]

    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

  38. [38]

    A comparative analysis of hybrid-quantum classical neural networks,

    K. Zamanet al., “A comparative analysis of hybrid-quantum classical neural networks,” inWorld Congress in Computer Science, Computer Engineering & Applied Computing, 2024

    work page 2024

  39. [39]

    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

  40. [40]

    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

  41. [41]

    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

  42. [42]

    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

  43. [43]

    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

  44. [44]

    Variational quantum algorithms,

    M. Cerezoet al., “Variational quantum algorithms,”Nature Reviews Physics, 2021

    work page 2021

  45. [45]

    Quantum circuit architecture search for variational quantum algorithms,

    Y . Duet al., “Quantum circuit architecture search for variational quantum algorithms,”npj Quantum Information, 2022

    work page 2022

  46. [46]

    Quantumdarts: differentiable quantum architecture search for variational quantum algorithms,

    W. Wuet al., “Quantumdarts: differentiable quantum architecture search for variational quantum algorithms,” inICML, 2023

    work page 2023

  47. [47]

    Self-supervised representation learning for bayesian quantum architecture search,

    Z. Heet al., “Self-supervised representation learning for bayesian quantum architecture search,”Physical Review A, 2025

    work page 2025

  48. [48]

    Noise-aware quantum architecture search based on nsga-ii algorithm,

    C. Li, H. Zeng, and D. Ding, “Noise-aware quantum architecture search based on nsga-ii algorithm,”arXiv:2601.10965, 2026

    work page arXiv 2026

  49. [49]

    Eqnas: Evolutionary quantum neural architecture search for image classification,

    Y . Liet al., “Eqnas: Evolutionary quantum neural architecture search for image classification,”Neural Networks, 2023

    work page 2023

  50. [50]

    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, 2025

    work page arXiv 2025

  51. [51]

    Closing the loop: Resource-aware hybrid nas guided by analytical and hardware-calibrated quantum cost modeling,

    M. Kashif, A. Marchisio, and M. Shafique, “Closing the loop: Resource- aware hybrid nas guided by analytical and hardware-calibrated quantum cost modeling,”arXiv:2603.00625, 2026

    work page arXiv 2026

  52. [52]

    Qnas: A neural architecture search framework for accurate and efficient quantum neural networks,

    K. Maleki, A. Marchisio, and M. Shafique, “Qnas: A neural architecture search framework for accurate and efficient quantum neural networks,” inIJCNN, 2026

    work page 2026

  53. [53]

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

    S. Duttaet al., “Qas-qtns: Curriculum reinforcement learning-driven quantum architecture search for quantum tensor networks,” inQCE, 2025

    work page 2025

  54. [54]

    Circuithunt: Automated quantum circuit screening for superior credit-card fraud detection,

    N. Innan, A. Singh, and M. Shafique, “Circuithunt: Automated quantum circuit screening for superior credit-card fraud detection,” inQAI, 2025

    work page 2025

  55. [55]

    Graph-based bayesian optimization for quantum circuit architecture search with uncertainty calibrated surrogates,

    P. K. Choudharyet al., “Graph-based bayesian optimization for quantum circuit architecture search with uncertainty calibrated surrogates,” arXiv:2512.09586, 2025

    work page arXiv 2025