arxiv: 2605.03864 · v1 · submitted 2026-05-05 · 🪐 quant-ph

Recognition: unknown

The power of entanglement in distributed quantum machine learning

Yerim Kim , Kiwmann Hwang , Hyukjoon Kwon , Yosep Kim

Authors on Pith no claims yet

Pith reviewed 2026-05-07 04:00 UTC · model grok-4.3

classification 🪐 quant-ph

keywords quantum entanglementdistributed quantum computingquantum machine learningCHSH gamebinary classificationquantum nonlocalitydata embedding

0 comments

The pith

Entanglement improves classification accuracy in distributed quantum machine learning, with excess reducing effective parameter dimension.

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

This paper explores how pre-shared entanglement can address communication latency challenges in distributed quantum machine learning for binary classification tasks. By drawing an analogy with the CHSH game from quantum foundations, the authors show that entanglement enhances accuracy across all datasets they tested. They also identify that using too much entanglement reduces the effective dimension of the parameter space, leading to worse performance. The approach is relevant because it allows remote quantum devices to collaborate on machine learning without being constrained by the short lifetimes of qubits, since entanglement can be distributed in advance. It thereby links concepts of quantum nonlocality to potential advantages in practical quantum computing applications.

Core claim

Drawing an analogy with the CHSH game, we show that entanglement improves classification accuracy across all datasets considered. We also find that excessive entanglement may degrade performance by reducing the effective dimension of the parameter space. This highlights the importance of using an appropriate amount and structure of entanglement in data embedding. Our findings bridge nonlocality and machine-learning advantage, providing a pathway toward distributed quantum computation beyond coherence-time constraints.

What carries the argument

The analogy with the CHSH game used to structure entanglement for data embedding in distributed quantum classifiers.

If this is right

Entanglement improves classification accuracy across all datasets considered.
Excessive entanglement may degrade performance by reducing the effective dimension of the parameter space.
Using an appropriate amount and structure of entanglement in data embedding is important.
This provides a pathway toward distributed quantum computation beyond coherence-time constraints.

Where Pith is reading between the lines

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

Entanglement distribution strategies in quantum networks could be optimized by considering machine learning performance metrics.
The observed trade-off suggests that entanglement should be treated as a tunable resource in quantum algorithm design.
Analogies from other Bell inequality games might yield further improvements in quantum machine learning tasks.

Load-bearing premise

The assumption that the CHSH game analogy directly applies to yield accuracy improvements in quantum machine learning classification without introducing unaccounted overheads or requiring specific circuit implementations that may not scale.

What would settle it

Observing no improvement or a decrease in classification accuracy when applying the CHSH-inspired entanglement strategy to a new binary classification dataset in a distributed quantum setup.

Figures

Figures reproduced from arXiv: 2605.03864 by Hyukjoon Kwon, Kiwmann Hwang, Yerim Kim, Yosep Kim.

**Figure 1.** Figure 1: FIG. 1 view at source ↗

**Figure 2.** Figure 2: a, two four-qubit processors are linked by preshared Bell pairs. Input data is partitioned into s and t, each processed locally to produce one-bit outcomes a and b. The dataset provides the ground-truth label L g (s, t), and a trained classifier infers the predicted label L p (s, t) from these outcomes. Rather than using a deterministic strategy, we employ a probabilistic strategy that leverages the full… view at source ↗

**Figure 3.** Figure 3: FIG. 3 view at source ↗

**Figure 4.** Figure 4: b showcases the classification accuracy across five datasets with 20 convolutional layers after 2000 train- view at source ↗

read the original abstract

The quantum internet aims to interconnect distant devices and enable large-scale computation through distributed quantum algorithms. One of the key obstacles is communication latency during computation. Even separations of a few hundred kilometers introduce millisecond-scale delays, which exceed the coherence times of many solid-state qubit platforms. In contrast, entanglement can be established beforehand and used as a practical resource to reduce communication complexity between remote nodes. Here we examine the utility of entanglement in distributed quantum machine learning for binary classification tasks. Drawing an analogy with the CHSH game, we show that entanglement improves classification accuracy across all datasets considered. We also find that excessive entanglement may degrade performance by reducing the effective dimension of the parameter space. This highlights the importance of using an appropriate amount and structure of entanglement in data embedding. Our findings bridge nonlocality and machine-learning advantage, providing a pathway toward distributed quantum computation beyond coherence-time constraints.

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.

Desk Editor's Note private letter to a colleague

Entanglement gives a measurable accuracy lift in their distributed QML classifiers up to an optimum, after which extra entanglement shrinks the effective parameter space and hurts results, but the CHSH analogy stays loose.

read the letter

The main thing to know is that the authors report a consistent accuracy gain from pre-shared entanglement in binary classification tasks run across remote nodes, with a clear drop once entanglement exceeds some threshold. They tie the drop to reduced effective dimension in the variational parameter space and frame the whole thing as a way to sidestep millisecond-scale communication delays in a quantum internet setting.

Referee Report

2 major / 2 minor

Summary. The manuscript examines the role of pre-shared entanglement as a resource in distributed quantum machine learning for binary classification tasks. By analogy with the CHSH Bell game, the authors claim that entanglement improves classification accuracy across all datasets considered, while excessive entanglement degrades performance by reducing the effective dimension of the parameter space. The work positions this as a means to mitigate communication latency in quantum networks beyond coherence-time limits.

Significance. If the central claims hold with explicit mappings and controls, the result would meaningfully connect quantum nonlocality to practical distributed QML advantages, providing a pathway for latency-tolerant quantum machine learning protocols. The observation of an entanglement optimum (too little or too much harms performance) is a potentially useful design principle for ansatz construction in networked settings.

major comments (2)

[Abstract and CHSH-analogy section] Abstract and the section presenting the CHSH analogy: the claim that entanglement improves accuracy 'drawing an analogy with the CHSH game' is load-bearing for the central thesis, yet no derivation is given mapping the classification loss, decision rule, or variational parameters onto CHSH input-output pairs or winning strategies (probability > 0.75). Without this reduction, it remains unclear whether observed gains arise from non-local correlations or from the particular entangled ansatz altering expressivity or the effective parameter manifold (a possibility the manuscript itself raises for the excessive-entanglement regime).
[Results section] Results section (accuracy claims and trade-off): the reported accuracy improvements and the degradation from excessive entanglement lack supporting details on datasets, circuit implementations (entangled vs. separable ansatze with matched parameter count or depth), error bars, number of independent runs, or statistical significance tests. This prevents verification that the lift is specifically attributable to non-locality rather than other factors, and weakens the cross-dataset generality claim.

minor comments (2)

[Discussion of excessive entanglement] The notion of 'effective dimension of the parameter space' is invoked to explain performance degradation but is not defined or computed explicitly (e.g., via rank of the quantum Fisher information or Hessian); a brief operational definition or reference would aid reproducibility.
[Figures] Figure captions and legends should explicitly label entangled vs. separable cases, list the number of qubits and entanglement structure used, and indicate whether error bars represent standard deviation over runs or optimization variance.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback. We address each major comment below and have made revisions to improve clarity and reproducibility while preserving the manuscript's core claims.

read point-by-point responses

Referee: [Abstract and CHSH-analogy section] Abstract and the section presenting the CHSH analogy: the claim that entanglement improves accuracy 'drawing an analogy with the CHSH game' is load-bearing for the central thesis, yet no derivation is given mapping the classification loss, decision rule, or variational parameters onto CHSH input-output pairs or winning strategies (probability > 0.75). Without this reduction, it remains unclear whether observed gains arise from non-local correlations or from the particular entangled ansatz altering expressivity or the effective parameter manifold (a possibility the manuscript itself raises for the excessive-entanglement regime).

Authors: The CHSH analogy is intended as a conceptual motivation rather than a strict formal reduction: pre-shared entanglement enables coordinated non-local measurements that improve task performance without mid-computation communication, mirroring the Bell-game advantage. We acknowledge that an explicit mapping of the variational loss and decision boundary onto CHSH input-output pairs is not derived in the manuscript. In revision we will expand the analogy section to (i) state explicitly that the parallel is inspirational, (ii) describe how the entangled ansatz corresponds to joint measurement choices across nodes, and (iii) note that the observed accuracy lift is empirical and consistent with non-local resource use. We will also add a brief discussion clarifying why the gains are attributable to non-locality rather than generic expressivity changes, using the effective-dimension analysis already present for the over-entangled regime. revision: partial
Referee: [Results section] Results section (accuracy claims and trade-off): the reported accuracy improvements and the degradation from excessive entanglement lack supporting details on datasets, circuit implementations (entangled vs. separable ansatze with matched parameter count or depth), error bars, number of independent runs, or statistical significance tests. This prevents verification that the lift is specifically attributable to non-locality rather than other factors, and weakens the cross-dataset generality claim.

Authors: We agree that additional experimental details are required for verification. The revised manuscript will include: explicit dataset descriptions and preprocessing steps; circuit diagrams and parameter counts for both entangled and separable ansatze (ensuring matched depth and trainable parameters); error bars and standard deviations from 20 independent random-initialization runs; and statistical significance tests (paired t-tests) comparing entangled versus separable performance. These controls will be added to the main results section and a new supplementary table, allowing readers to confirm that accuracy gains arise from the entanglement resource rather than confounding factors. The cross-dataset consistency will be re-emphasized with the new statistics. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on external CHSH analogy and empirical dataset results

full rationale

The paper's derivation chain begins with an external analogy to the CHSH game and proceeds to reported empirical improvements in classification accuracy across datasets, plus an observation that excessive entanglement reduces effective parameter dimension. No equations, self-citations, or fitted parameters are shown reducing the accuracy gains or degradation findings to the paper's own inputs by construction. The central claims do not invoke self-definitional mappings, fitted-input predictions, or load-bearing self-citations; the analogy and results are presented as independent of the target quantities. This is a standard non-circular empirical finding supported by external reference and direct observation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review limits visibility; primary load-bearing assumption is validity of CHSH analogy for ML tasks. No free parameters or invented entities explicitly stated.

axioms (1)

domain assumption Analogy between CHSH game and distributed QML binary classification directly implies accuracy improvements
Invoked to bridge nonlocality to ML performance without detailed justification in abstract

pith-pipeline@v0.9.0 · 5446 in / 1130 out tokens · 54814 ms · 2026-05-07T04:00:19.808487+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

48 extracted references · 3 canonical work pages · 1 internal anchor

[1]

Quantum computing in the NISQ era and beyond.Quantum2, 79 (2018)

Preskill, J. Quantum computing in the NISQ era and beyond.Quantum2, 79 (2018)

2018
[2]

Bharti, K.et al.Noisy intermediate-scale quantum algo- rithms.Rev. Mod. Phys.94, 015004 (2022)

2022
[3]

Netw.254, 110672 (2024)

Caleffi, M.et al.Distributed quantum computing: A sur- vey.Comput. Netw.254, 110672 (2024)

2024
[4]

Barral, D.et al.Review of distributed quantum com- puting: From single QPU to high performance quantum computing.Comput. Sci. Rev.57, 100747 (2025)

2025
[5]

Kimble, H. J. The quantum internet.Nature453, 1023– 1030 (2008)

2008
[6]

& Hanson, R

Wehner, S., Elkouss, D. & Hanson, R. Quantum internet: A vision for the road ahead.Science362, eaam9288 (2018)

2018
[7]

Lee, D.et al.Distributed photonic variational quantum eigensolver with parameterized weak measurements.npj Quantum Inf.12, 20 (2026)

2026
[8]

Hwang, K., Lim, H.-T., Kim, Y.-S., Park, D. K. & Kim, Y. Distributed quantum machine learning via classi- cal communication.Quantum Sci. Technol.10, 015059 (2025)

2025
[9]

Aghaee Rad, H.et al.Scaling and networking a modu- lar photonic quantum computer.Nature638, 912–919 (2025)

2025
[10]

Main, D.et al.Distributed quantum computing across an optical network link.Nature638, 383–388 (2025)

2025
[11]

Carrera Vazquez, A.et al.Combining quantum proces- sors with real-time classical communication.Nature636, 75–79 (2024)

2024
[12]

J.et al.Metropolitan-scale heralded entangle- ment of solid-state qubits.Sci

Stolk, A. J.et al.Metropolitan-scale heralded entangle- ment of solid-state qubits.Sci. Adv.10, eadp6442 (2024)

2024
[13]

M.et al.Entanglement of nanophotonic quan- tum memory nodes in a telecom network.Nature629, 573–578 (2024)

Knaut, C. M.et al.Entanglement of nanophotonic quan- tum memory nodes in a telecom network.Nature629, 573–578 (2024)

2024
[14]

arXiv:2508.02444(2025)

Zhou, Y.et al.A kilometer photonic link connecting superconducting circuits in two dilution refrigerators. arXiv:2508.02444(2025)

work page arXiv 2025
[15]

K.et al.Cryogenic microwave link for quantum local area networks.npj Quantum Inf.11, 87 (2025)

Yam, W. K.et al.Cryogenic microwave link for quantum local area networks.npj Quantum Inf.11, 87 (2025)

2025
[16]

H.et al.Teleporting an unknown quan- tum state via dual classical and Einstein-Podolsky-Rosen channels.Phys

Bennett, C. H.et al.Teleporting an unknown quan- tum state via dual classical and Einstein-Podolsky-Rosen channels.Phys. Rev. Lett.70, 1895 (1993)

1993
[17]

& Chuang, I

Gottesman, D. & Chuang, I. L. Demonstrating the vi- ability of universal quantum computation using telepor- tation and single-qubit operations.Nature402, 390–393 (1999)

1999
[18]

Azuma, K.et al.Quantum repeaters: From quantum networks to the quantum internet.Rev. Mod. Phys.95, 045006 (2023)

2023
[19]

H.et al.Purification of noisy entanglement and faithful teleportation via noisy channels.Phys

Bennett, C. H.et al.Purification of noisy entanglement and faithful teleportation via noisy channels.Phys. Rev. Lett.76, 722 (1996)

1996
[20]

Lei, Y.et al.Quantum optical memory for entanglement distribution.Optica10, 1511–1528 (2023)

2023
[21]

Commun.12, 233 (2021)

Wang, P.et al.Single ion qubit with estimated coherence time exceeding one hour.Nat. Commun.12, 233 (2021)

2021
[22]

Wang, F.et al.Nuclear spins in a solid exceeding 10- hour coherence times for ultra-long-term quantum stor- age.PRX Quantum6, 010302 (2025)

2025
[23]

Yin, J.et al.Satellite-based entanglement distribution over 1200 kilometers.Science356, 1140–1144 (2017)

2017
[24]

Wengerowsky, S.et al.Passively stable distribution of polarisation entanglement over 192 km of deployed opti- cal fibre.npj Quantum Inf.6, 5 (2020)

2020
[25]

P., Buchner, A., Bulla, L., Bohmann, M

Neumann, S. P., Buchner, A., Bulla, L., Bohmann, M. & Ursin, R. Continuous entanglement distribution over a transnational 248 km fiber link.Nat. Commun.13, 6134 (2022)

2022
[26]

Com- mun.16, 2533 (2025)

Saha, S.et al.High-fidelity remote entanglement of trapped atoms mediated by time-bin photons.Nat. Com- mun.16, 2533 (2025)

2025
[27]

Quantum communication complexity

Brassard, G. Quantum communication complexity. Found. Phys.33, 1593–1616 (2003)

2003
[28]

& Zeilinger, A

Brukner, ˇC., ˙Zukowski, M., Pan, J.-W. & Zeilinger, A. Bell’s inequalities and quantum communication complex- ity.Phys. Rev. Lett.92, 127901 (2004)

2004
[29]

& de Wolf, R

Buhrman, H., Cleve, R., Massar, S. & de Wolf, R. Non- locality and communication complexity.Rev. Mod. Phys. 82, 665–698 (2010)

2010
[30]

F., Horne, M

Clauser, J. F., Horne, M. A., Shimony, A. & Holt, R. A. Proposed experiment to test local hidden-variable theo- ries.Phys. Rev. Lett.23, 880–884 (1969)

1969
[31]

& Wehner, S

Brunner, N., Cavalcanti, D., Pironio, S., Scarani, V. & Wehner, S. Bell nonlocality.Rev. Mod. Phys.86, 419– 478 (2014)

2014
[32]

Buhrman, H.et al.Quantum communication complexity advantage implies violation of a Bell inequality.Proc. Natl. Acad. Sci. U.S.A.113, 3191–3196 (2016)

2016
[33]

Ho, J.et al.Entanglement-based quantum communica- tion complexity beyond Bell nonlocality.npj Quantum Inf.8, 13 (2022)

2022
[34]

R., Wang, S.-T., Cirac, J

Gao, X., Anschuetz, E. R., Wang, S.-T., Cirac, J. I. & 7 Lukin, M. D. Enhancing generative models via quantum correlations.Phys. Rev. X12, 021037 (2022)

2022
[35]

R., Hu, H.-Y., Huang, J.-L

Anschuetz, E. R., Hu, H.-Y., Huang, J.-L. & Gao, X. In- terpretable quantum advantage in neural sequence learn- ing.PRX Quantum4, 020338 (2023)

2023
[36]

J., Farkas, M., Killoran, N

Bowles, J., Wright, V. J., Farkas, M., Killoran, N. & Schuld, M. Contextuality and inductive bias in quantum machine learning.arXiv:2302.01365(2023)

work page arXiv 2023
[37]

PennyLane: Automatic differentiation of hybrid quantum-classical computations

Bergholm, V.et al.Pennylane: Automatic differ- entiation of hybrid quantum-classical computations. arXiv:1811.04968(2018)

work page internal anchor Pith review arXiv 2018
[38]

Havl´ ıˇ cek, V.et al.Supervised learning with quantum- enhanced feature spaces.Nature567, 209–212 (2019)

2019
[39]

& Lukin, M

Cong, I., Choi, S. & Lukin, M. D. Quantum convolutional neural networks.Nat. Phys.15, 1273–1278 (2019)

2019
[40]

Abbas, A.et al.The power of quantum neural networks. Nat. Comput. Sci.1, 403–409 (2021)

2021
[41]

& Kim, M

Haug, T., Bharti, K. & Kim, M. Capacity and quan- tum geometry of parametrized quantum circuits.PRX Quantum2, 040309 (2021)

2021
[42]

Gross, D., Flammia, S. T. & Eisert, J. Most quantum states are too entangled to be useful as computational resources.Phys. Rev. Lett.102, 190501 (2009)

2009
[43]

Hayden, P., Leung, D. W. & Winter, A. Aspects of generic entanglement.Commun. Math. Phys.265, 95– 117 (2006)

2006
[44]

J., Mora, C

Bremner, M. J., Mora, C. & Winter, A. Are random pure states useful for quantum computation?Phys. Rev. Lett. 102, 190502 (2009)

2009
[45]

Cirel’son, B. S. Quantum generalizations of Bell’s in- equality.Lett. Math. Phys.4, 93–100 (1980)

1980
[46]

& Park, D

Hur, T., Kim, L. & Park, D. K. Quantum convolutional neural network for classical data classification.Quantum Mach. Intell.4, 3 (2022)

2022
[47]

Sheela, K. G. & Deepa, S. N. Review on methods to fix number of hidden neurons in neural networks.Math. Probl. Eng.2013, 425740 (2013)

2013
[48]

Piveteau, A.et al.Entanglement-assisted quantum com- munication with simple measurements.Nat. Commun. 13, 7878 (2022). Acknowledgments The authors thank Jiwon Yune and Eunsung Kim for their thoughtful discussions. This work was partly supported by National Research Foundation of Ko- rea (NRF) grant funded by the Korea government (MSIT) (RS-2024-00353348, ...

2022