Pith. sign in

REVIEW 1 major objections 1 minor 41 references

MZeQAS finds high-performing variational quantum circuit architectures faster by estimating performance from QNTK Gram matrix convergence without training candidates.

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 →

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

T0 review · grok-4.3

2026-06-30 22:14 UTC pith:UK2GWDW4

load-bearing objection MZeQAS puts a QNTK Gram-matrix convergence proxy into MCTS for zero-shot VQA architecture search, which is new but rests on an unverified convergence claim. the 1 major comments →

arxiv 2605.27410 v2 pith:UK2GWDW4 submitted 2026-05-12 quant-ph cs.LGcs.NE

Zero-shot Quantum Neural Architecture Search

Tung Dao , Son N. Tran , Huynh Thi Thanh Binh This is my paper
classification quant-ph cs.LGcs.NE
keywords quantum neural architecture searchzero-shot surrogatequantum neural tangent kernelvariational quantum algorithmsmonte carlo tree searchNISQ devices
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 paper identifies a setting where the Gram matrix of the Quantum Neural Tangent Kernel converges and uses that convergence to build a zero-shot surrogate that predicts how well a candidate circuit will perform after training. This surrogate replaces the expensive step of fully training every circuit during the search, allowing an MCTS-based algorithm called MZeQAS to explore the space of possible architectures more efficiently. The authors report that the resulting circuits outperform those found by earlier evolutionary quantum neural architecture search methods both in final solution quality and in the computational cost of the search itself. A reader would care because the cost of repeatedly training candidate circuits has been a practical barrier to deploying variational quantum algorithms on near-term hardware.

Core claim

The authors identify a convergence setting for the Gram matrix of the Quantum Neural Tangent Kernel and show that this property supplies a zero-shot surrogate capable of ranking candidate circuits by expected performance; when this surrogate is combined with Monte Carlo Tree Search, the resulting MZeQAS framework produces higher-quality variational quantum circuits than existing evolutionary methods while using substantially less search time.

What carries the argument

The zero-shot surrogate model constructed from the converged QNTK Gram matrix, which supplies performance estimates that replace full training of each candidate circuit inside the MCTS search.

Load-bearing premise

The Gram matrix of the Quantum Neural Tangent Kernel converges for the candidate circuits that appear in the architecture search.

What would settle it

Measure the actual trained performance of a set of circuits and compare it to the ranking produced by the QNTK Gram-matrix surrogate; if the surrogate ranking shows no correlation with the trained results on the circuits used in the search, the method fails.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Architecture search for variational quantum algorithms no longer requires training every candidate circuit to completion.
  • The computational budget for searching over circuit structures can be spent on a larger number of candidates or on deeper exploration.
  • Higher-performing circuits become reachable within the same search-time limits that previously restricted evolutionary methods.
  • The same proxy estimation step can be reused across multiple VQA tasks once the convergence property holds.

Where Pith is reading between the lines

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

  • If the convergence condition holds more broadly than the tested instances, the surrogate could be dropped into other quantum circuit optimization loops that currently rely on full retraining.
  • The approach opens a route to treating quantum architecture search more like classical zero-shot neural architecture search, where cheap proxies replace expensive evaluations.
  • Hardware-specific constraints such as gate error rates could be folded into the same Gram-matrix calculation to produce device-aware rankings without additional training runs.

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

1 major / 1 minor

Summary. The manuscript proposes MZeQAS, an MCTS-based zero-shot quantum neural architecture search framework for variational quantum algorithms. It claims to identify a setting in which the Gram matrix of the Quantum Neural Tangent Kernel converges, enabling a zero-shot surrogate to rank candidate circuits without full training, and reports that this yields superior search efficiency and solution quality compared to existing evolutionary QNAS methods.

Significance. If the claimed QNTK Gram-matrix convergence holds with verifiable scope conditions and the surrogate correlates with post-training performance, the method could meaningfully reduce the repeated-training cost that currently limits evolutionary QNAS, offering a more scalable route to architecture design for NISQ-era VQAs. The combination of a kernel-based proxy with MCTS exploration is a technically coherent idea whose practical value hinges on the surrogate's fidelity.

major comments (1)
  1. [Abstract] Abstract (and central claim): the zero-shot surrogate is justified by the assertion that 'a setting in which the Gram matrix of the Quantum Neural Tangent Kernel converges' has been identified, yet the manuscript supplies neither explicit scope conditions (circuit depth, parameter regime, ansatz family) nor a derivation or theorem establishing convergence. Because the surrogate is used to rank candidates without any training, this unproven convergence is load-bearing for all reported performance gains.
minor comments (1)
  1. The abstract would be strengthened by a single sentence stating the concrete circuit families or depth bounds under which the Gram-matrix convergence was observed.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and for highlighting the need for stronger justification of the central convergence claim. We address the major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract (and central claim): the zero-shot surrogate is justified by the assertion that 'a setting in which the Gram matrix of the Quantum Neural Tangent Kernel converges' has been identified, yet the manuscript supplies neither explicit scope conditions (circuit depth, parameter regime, ansatz family) nor a derivation or theorem establishing convergence. Because the surrogate is used to rank candidates without any training, this unproven convergence is load-bearing for all reported performance gains.

    Authors: We agree that the manuscript would be strengthened by explicit scope conditions and a formal derivation. In the revision we will add a new subsection (likely in Section 3) that states the precise conditions on circuit depth, parameter initialization regime, and ansatz family under which the QNTK Gram-matrix convergence is observed, together with a derivation based on the Lipschitz continuity of the quantum feature map and the resulting contraction of the kernel matrix. We will also include a short proof sketch and additional numerical verification across the stated regimes. These additions will make the load-bearing assumption fully explicit and verifiable. revision: yes

Circularity Check

0 steps flagged

No circularity: central claim rests on empirical identification of convergence regime, not self-referential derivation

full rationale

The abstract states that the authors 'identify a setting in which the Gram matrix of the Quantum Neural Tangent Kernel converges' and then 'design a zero-shot surrogate model' based on that observation. No equations, self-citations, or fitted parameters are shown that would reduce the surrogate performance estimate to the identification step by construction. The reported outperformance is presented as an experimental result using the surrogate inside MCTS, with no load-bearing step that renames a fit as a prediction or imports uniqueness from prior self-work. Because the provided text contains no derivation chain that collapses to its inputs, the finding is no significant circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on an asserted convergence property of the QNTK Gram matrix whose scope is not delimited in the abstract; no free parameters, axioms, or invented entities are explicitly listed because the full text is unavailable.

pith-pipeline@v0.9.1-grok · 5735 in / 1114 out tokens · 15844 ms · 2026-06-30T22:14:28.138660+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Zero-shot Quantum Neural Architecture Search." pith.science (2026). https://pith.science/paper/UK2GWDW4

@misc{pith2026260527410,
  author       = {Pith},
  title        = {Pith review of: Zero-shot Quantum Neural Architecture Search},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UK2GWDW4}},
  note         = {Machine review of arXiv:2605.27410}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Variational Quantum Algorithms (VQAs) are a leading approach to exploiting near-term quantum hardware, leveraging parameterized quantum circuits and classical optimization to achieve advantage. Despite their promise, the practical deployment of VQAs is challenged by the difficulty of designing quantum circuit architectures that balance expressivity, trainability, and hardware constraints. Existing evolutionary-based quantum neural architecture search methods address these challenges but suffer from high computational costs due to repeated training of candidate circuits. In this work, we identify a setting in which the Gram matrix of the Quantum Neural Tangent Kernel converges. Building on this observation, we design a zero-shot surrogate model to estimate candidate performance without full training, significantly accelerating the architecture search process. Using this surrogate, we propose MZeQAS, a Monte Carlo Tree Search (MCTS)-based Zero-Shot Quantum Neural Architecture Search framework for VQAs. By integrating proxy-based performance estimation with MCTS exploration, MZeQAS efficiently discovers high-performing architectures. Experimental results demonstrate that MZeQAS outperforms existing approaches in terms of both search efficiency and solution quality, providing a scalable and effective framework for advancing VQA deployment on noisy intermediate-scale quantum devices.

Figures

Figures reproduced from arXiv: 2605.27410 by Huynh Thi Thanh Binh, Son N. Tran, Tung Dao.

Figure 1
Figure 1. Figure 1: Each iteration of MZeQAS consists of four stages, from left to right: [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Correlation analysis between train and test losses against AMES in [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The topology IBM’s 5 qubits device ibmq_quito [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗

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

41 extracted references · 41 canonical work pages · 1 internal anchor

  1. [1]

    Variational quantum algorithms.Nature Reviews Physics, 3(9):625–644, 2021

    Marco Cerezo, Andrew Arrasmith, Ryan Babbush, Simon C Benjamin, Suguru Endo, Keisuke Fujii, Jarrod R McClean, Kosuke Mitarai, Xiao Yuan, Lukasz Cincio, et al. Variational quantum algorithms.Nature Reviews Physics, 3(9):625–644, 2021

    work page 2021

  2. [2]

    Quantum computing in the nisq era and beyond.Quantum, 2:79, 2018

    John Preskill. Quantum computing in the nisq era and beyond.Quantum, 2:79, 2018

    work page 2018

  3. [3]

    Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets

    Abhinav Kandala, Antonio Mezzacapo, Kristan Temme, Maika Takita, Markus Brink, Jerry M Chow, and Jay M Gambetta. Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets. nature, 549(7671):242–246, 2017

    work page 2017

  4. [4]

    Generalized unitary coupled cluster wave functions for quantum computation.Journal of chemical theory and computation, 15(1):311–324, 2018

    Joonho Lee, William J Huggins, Martin Head-Gordon, and K Birgitta Whaley. Generalized unitary coupled cluster wave functions for quantum computation.Journal of chemical theory and computation, 15(1):311–324, 2018

    work page 2018

  5. [5]

    Quantum neural architecture search with quantum circuits metric and bayesian optimization

    Trong Duong, Sang T Truong, Minh Tam, Bao Bach, Ju-Young Ryu, and June-Koo Kevin Rhee. Quantum neural architecture search with quantum circuits metric and bayesian optimization. InICML 2022 2nd AI for Science Workshop, 2022

    work page 2022

  6. [6]

    Neural predictor based quantum architecture search.Machine Learning: Science and Technology, 2(4):045027, oct 2021

    Shi-Xin Zhang, Chang-Yu Hsieh, Shengyu Zhang, and Hong Yao. Neural predictor based quantum architecture search.Machine Learning: Science and Technology, 2(4):045027, oct 2021. doi: 10.1088/2632-2153/ac28dd. URLhttps://doi.org/10.1088/2632-2153/ac28dd

    work page doi:10.1088/2632-2153/ac28dd 2021

  7. [7]

    Dif- ferentiable quantum architecture search.Quantum Science and Tech- nology, 7(4):045023, aug 2022

    Shi-Xin Zhang, Chang-Yu Hsieh, Shengyu Zhang, and Hong Yao. Dif- ferentiable quantum architecture search.Quantum Science and Tech- nology, 7(4):045023, aug 2022. doi: 10.1088/2058-9565/ac87cd. URL https://doi.org/10.1088/2058-9565/ac87cd

    work page doi:10.1088/2058-9565/ac87cd 2022

  8. [8]

    Quantum architecture search via deep reinforcement learning,

    En-Jui Kuo, Yao-Lung L. Fang, and Samuel Yen-Chi Chen. Quantum architecture search via deep reinforcement learning, 2021. URLhttps: //arxiv.org/abs/2104.07715

    work page arXiv 2021

  9. [9]

    Quantumnas: Noise-adaptive search for robust quan- tum circuits

    Hanrui Wang, Yongshan Ding, Jiaqi Gu, Yujun Lin, David Z Pan, Frederic T Chong, and Song Han. Quantumnas: Noise-adaptive search for robust quan- tum circuits. In2022 IEEE International Symposium on High-Performance Computer Architecture (HPCA), pages 692–708. IEEE, 2022. 14

    work page 2022

  10. [10]

    Adaptive quantum approximate optimization algorithm for solving combinatorial problems on a quantum computer.Physical Review Research, 4(3):033029, 2022

    Linghua Zhu, Ho Lun Tang, George S Barron, FA Calderon-Vargas, Nicholas J Mayhall, Edwin Barnes, and Sophia E Economou. Adaptive quantum approximate optimization algorithm for solving combinatorial problems on a quantum computer.Physical Review Research, 4(3):033029, 2022

    work page 2022

  11. [11]

    Eqnas: Evolutionary quantum neural architecture search for image classification.Neural Networks, 168:471–483, 2023

    Yangyang Li, Ruijiao Liu, Xiaobin Hao, Ronghua Shang, Peixiang Zhao, and Licheng Jiao. Eqnas: Evolutionary quantum neural architecture search for image classification.Neural Networks, 168:471–483, 2023

    work page 2023

  12. [12]

    Quantum circuit design using a progressive widening enhanced monte carlo tree search.Advanced Quantum Technologies, page e2500093, 2025

    Vincenzo Lipardi, Domenica Dibenedetto, Georgios Stamoulis, and Mark HM Winands. Quantum circuit design using a progressive widening enhanced monte carlo tree search.Advanced Quantum Technologies, page e2500093, 2025

    work page 2025

  13. [13]

    Neural predictor based quantum architecture search.Machine Learning: Science and Technology, 2(4):045027, 2021

    Shi-Xin Zhang, Chang-Yu Hsieh, Shengyu Zhang, and Hong Yao. Neural predictor based quantum architecture search.Machine Learning: Science and Technology, 2(4):045027, 2021

    work page 2021

  14. [14]

    Training-free quantum architecture search

    Zhimin He, Maijie Deng, Shenggen Zheng, Lvzhou Li, and Haozhen Situ. Training-free quantum architecture search. InProceedings of the AAAI conference on artificial intelligence, volume 38, pages 12430–12438, 2024

    work page 2024

  15. [15]

    Automated quantum circuit design with nested monte carlo tree search.IEEE Transactions on Quantum Engineering, 4:1–20, 2023

    PeiyongWang, MuhammadUsman, UdayaParampalli, LloydCLHollenberg, and Casey R Myers. Automated quantum circuit design with nested monte carlo tree search.IEEE Transactions on Quantum Engineering, 4:1–20, 2023

    work page 2023

  16. [16]

    A meta-trained generator for quantum architecture search.EPJ Quantum Technology, 11(1):44, 2024

    Zhimin He, Chuangtao Chen, Zhengjiang Li, Haozhen Situ, Fei Zhang, Shenggen Zheng, and Lvzhou Li. A meta-trained generator for quantum architecture search.EPJ Quantum Technology, 11(1):44, 2024

    work page 2024

  17. [17]

    Quantum architecture search with neural predictor based on graph measures.Advanced Quantum Technologies, 7(11):2400223, 2024

    Zhimin He, Zhengjiang Li, Maijie Deng, Shenggen Zheng, Haozhen Situ, and Lvzhou Li. Quantum architecture search with neural predictor based on graph measures.Advanced Quantum Technologies, 7(11):2400223, 2024

    work page 2024

  18. [18]

    Barren plateaus in variational quantum computing

    Martin Larocca, Supanut Thanasilp, Samson Wang, Kunal Sharma, Jacob Biamonte, Patrick J Coles, Lukasz Cincio, Jarrod R McClean, Zoë Holmes, and Marco Cerezo. Barren plateaus in variational quantum computing. Nature Reviews Physics, pages 1–16, 2025

    work page 2025

  19. [19]

    Zen-nas: A zero-shot nas for high-performance image recognition

    Minghao Lin, Ping Wang, Zhenhong Sun, Hao Chen, Xiaohui Sun, Qinghua Qian, Hong Li, and Rong Jin. Zen-nas: A zero-shot nas for high-performance image recognition. InProceedings of the IEEE/CVF International Confer- ence on Computer Vision (ICCV), pages 337–346, October 2021

    work page 2021

  20. [20]

    Meco: zero-shot nas with one data and single forward pass via minimum eigenvalue of correlation

    Tangyu Jiang, Haodi Wang, and Rongfang Bie. Meco: zero-shot nas with one data and single forward pass via minimum eigenvalue of correlation. Advances in Neural Information Processing Systems, 36:61020–61047, 2023. 15

    work page 2023

  21. [21]

    Zico: Zero-shot nas via inverse coefficient of variation on gradients

    Guihong Li, Yuedong Yang, Kartikeya Bhardwaj, and Radu Marculescu. Zico: Zero-shot nas via inverse coefficient of variation on gradients. InThe Eleventh International Conference on Learning Representations, 2023

    work page 2023

  22. [22]

    Parzc: Parametric zero-cost proxies for efficient nas

    Peijie Dong, Lujun Li, Zhenheng Tang, Xiang Liu, Zimian Wei, Qiang Wang, and Xiaowen Chu. Parzc: Parametric zero-cost proxies for efficient nas. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 16327–16335, 2025

    work page 2025

  23. [23]

    Neural architecture search on imagenet in four gpu hours: A theoretically inspired perspective

    Wuyang Chen, Xinyu Gong, and Zhangyang Wang. Neural architecture search on imagenet in four gpu hours: A theoretically inspired perspective. InInternational Conference on Learning Representations (ICLR), 2021

    work page 2021

  24. [24]

    Uni- fying and boosting gradient-based training-free neural architecture search

    Yao Shu, Zhongxiang Dai, Zhaoxuan Wu, and Bryan Kian Hsiang Low. Uni- fying and boosting gradient-based training-free neural architecture search. Advances in neural information processing systems, 35:33001–33015, 2022

    work page 2022

  25. [25]

    Gen- eralization properties of nas under activation and skip connection search

    Zhenyu Zhu, Fanghui Liu, Grigorios Chrysos, and Volkan Cevher. Gen- eralization properties of nas under activation and skip connection search. Advances in Neural Information Processing Systems, 35:23551–23565, 2022

    work page 2022

  26. [26]

    Du, Xiyu Zhai, Barnabás Póczos, and Aarti Singh

    Simon S. Du, Xiyu Zhai, Barnabás Póczos, and Aarti Singh. Gradient descent provably optimizes over-parameterized neural networks. In7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net, 2019. URLhttps: //openreview.net/forum?id=S1eK3i09YQ

    work page 2019

  27. [27]

    An- alyzing convergence in quantum neural networks: deviations from neural tangent kernels

    Xuchen You, Shouvanik Chakrabarti, Boyang Chen, and Xiaodi Wu. An- alyzing convergence in quantum neural networks: deviations from neural tangent kernels. InInternational Conference on Machine Learning, pages 40199–40224. PMLR, 2023

    work page 2023

  28. [28]

    Neural tangent kernel: Convergence and generalization in neural networks

    Arthur Jacot, Franck Gabriel, and Clément Hongler. Neural tangent kernel: Convergence and generalization in neural networks. InNeurIPS, 2018

    work page 2018

  29. [29]

    MIT Press, 2016.http://www.deeplearningbook.org

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville.Deep Learning. MIT Press, 2016.http://www.deeplearningbook.org

    work page 2016

  30. [30]

    Monte carlo tree search for comprehensive exploration in llm-based automatic heuristic design

    Zhi Zheng, Zhuoliang Xie, Zhenkun Wang, and Bryan Hooi. Monte carlo tree search for comprehensive exploration in llm-based automatic heuristic design. InForty-second International Conference on Machine Learning, 2025

    work page 2025

  31. [31]

    elo ratings

    Rémi Coulom. Computing “elo ratings” of move patterns in the game of go. ICGA journal, 30(4):198–208, 2007

    work page 2007

  32. [32]

    The mnist database of handwritten digit images for machine learning research [best of the web].IEEE signal processing magazine, 29(6): 141–142, 2012

    Li Deng. The mnist database of handwritten digit images for machine learning research [best of the web].IEEE signal processing magazine, 29(6): 141–142, 2012. 16

    work page 2012

  33. [33]

    Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

    Han Xiao, Kashif Rasul, and Roland Vollgraf. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms.arXiv preprint arXiv:1708.07747, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  34. [34]

    Connection- ist bench (vowel recognition - deterding data)

    David Deterding, Mahesan Niranjan, and Tony Robinson. Connection- ist bench (vowel recognition - deterding data). UCI Machine Learning Repository, 1988. DOI: https://doi.org/10.24432/C58P4S

    work page doi:10.24432/c58p4s 1988

  35. [35]

    Quantum phase transitions.Physics world, 12(4):33, 1999

    Subir Sachdev. Quantum phase transitions.Physics world, 12(4):33, 1999

    work page 1999

  36. [36]

    Ma- chine learning quantum phases of matter beyond the fermion sign problem

    Peter Broecker, Juan Carrasquilla, Roger G Melko, and Simon Trebst. Ma- chine learning quantum phases of matter beyond the fermion sign problem. Scientific reports, 7(1):8823, 2017

    work page 2017

  37. [37]

    Machine learning phases of matter

    Juan Carrasquilla and Roger G Melko. Machine learning phases of matter. Nature Physics, 13(5):431–434, 2017

    work page 2017

  38. [38]

    Quantum phases of matter on a 256-atom programmable quantum simulator.Nature, 595(7866):227–232, 2021

    Sepehr Ebadi, Tout T Wang, Harry Levine, Alexander Keesling, Giulia Semeghini, Ahmed Omran, Dolev Bluvstein, Rhine Samajdar, Hannes Pichler, Wen Wei Ho, et al. Quantum phases of matter on a 256-atom programmable quantum simulator.Nature, 595(7866):227–232, 2021

    work page 2021

  39. [39]

    Cambridge University Press, 2023

    Subir Sachdev.Quantum phases of matter. Cambridge University Press, 2023

    work page 2023

  40. [40]

    Understanding generalization in quantum machine learning with margins

    Tak Hur and Daniel K Park. Understanding generalization in quantum machine learning with margins. InForty-second International Conference on Machine Learning, 2025

    work page 2025

  41. [41]

    Quantum-train: Rethinking hybrid quantum-classical machine learning in the model com- pression perspective.Quantum Machine Intelligence, 7(2):80, 2025

    Chen-Yu Liu, En-Jui Kuo, Chu-Hsuan Abraham Lin, Jason Gemsun Young, Yeong-Jar Chang, Min-Hsiu Hsieh, and Hsi-Sheng Goan. Quantum-train: Rethinking hybrid quantum-classical machine learning in the model com- pression perspective.Quantum Machine Intelligence, 7(2):80, 2025. 17 A Additional numerical results Table 5: Comparison between GA with weight sharing...

    work page 2025