Pith. sign in

REVIEW 3 major objections 4 minor 45 references

The paper claims that machine learning can identify and quantify quantum entanglement directly from raw, noise-corrupted SIC-POVM photon counts, without density-matrix reconstruction or conventional separability criteria.

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 · deepseek-v4-flash

2026-08-01 04:21 UTC pith:TC2D2VZW

load-bearing objection A careful proof of concept that ML can screen two-qubit entanglement from simulated SIC-POVM counts; the body is honest about limits, the abstract less so. the 3 major comments →

arxiv 2607.22853 v1 pith:TC2D2VZW submitted 2026-07-24 quant-ph physics.comp-phphysics.data-anphysics.optics

Machine Learning of Quantum Entanglement from Noisy Measurements

Artur Czerwinski , Maciej Wi\'sniewski , Pawe{\l} Moszczy\'nski This is my paper
classification quant-ph physics.comp-phphysics.data-anphysics.optics PACS 03.67.Mn42.50.Ex07.05.Mh
keywords quantum entanglementmachine learningSIC-POVMquantum state tomographyconcurrenceWerner statesphoton countingPoisson noise
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 argues that entanglement information is already fully present in the raw, noise-corrupted outcomes of a SIC-POVM measurement, so machine-learning models can classify and quantify entanglement without tomography. Using simulated two-qubit polarization states and Poissonian photon-counting noise, it shows that separable and Bell states are almost perfectly separated, Werner states are classified with roughly 0.995 accuracy, and concurrence is regressed with R² near 0.999. This matters because it suggests a fast, direct route from experimental counts to entanglement verdicts, bypassing density-matrix reconstruction. The authors themselves frame the result as a proof of concept valid under Poissonian counting noise, not yet under full experimental imperfections.

Core claim

The central claim is that the 16 coincidence counts obtained from a two-qubit SIC-POVM, even when Poisson-noisy, carry enough information for ML models to learn entanglement properties directly. In the separable-versus-Bell dataset, KNN, SVC, and Random Forest reach 100% accuracy with only 40 training examples, and KMeans clusters the 20,000 vectors into the two physical classes with only 5 errors. For Werner states, SVC classifies separable vs entangled with accuracy 0.99498, with errors concentrated at the separability boundary p=1/3; Random Forest regression estimates concurrence with test R²=0.99947 and MAE=0.00315, outperforming a QST baseline. The paper concludes that entanglement stru

What carries the argument

The load-bearing object is the 16-dimensional SIC-POVM count vector: a symmetric, informationally complete measurement whose 16 outcomes give a tomographically complete fingerprint of the two-qubit state. For each quantum state, Born probabilities are computed and converted to Poisson-distributed coincidence counts, producing a noisy fingerprint of the state. This vector serves simultaneously as input for supervised classifiers, unsupervised clustering, and regression. Diversity is added by local-unitary rotations of Bell, separable, and Werner states, which preserve entanglement while changing the measurement statistics. The workhorse methods are kernel SVC, KNN, Random Forest, and SVR/RFR,

Load-bearing premise

The results rest on the assumption that simulated Poissonian photon-counting noise faithfully represents real experimental noise; if real measurements include losses, dark counts, dead time, or other distortions, the reported classification and regression accuracies are not guaranteed to transfer.

What would settle it

Run the same pipeline on real SIC-POVM counts from a polarization-entangled photon source and compare ML predictions to full tomography; or rerun the simulations with detector efficiency below 100% and dark counts added. If accuracy on Werner states near the separability boundary or regression R² drops sharply, the claim that raw counts alone carry enough information for ML under realistic noise would be falsified.

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

If this is right

  • In the Poisson-noise regime studied, a 16-dimensional SIC-POVM count vector is sufficient for high-accuracy binary entanglement classification: near-perfect accuracy holds with as few as 40 training samples for well-separated classes.
  • Werner-state classification errors concentrate at the separability boundary p=1/3, so ML confidence tracks physical distance from the separable-entangled transition.
  • Random Forest regression estimates concurrence from raw counts more accurately than least-squares QST followed by analytic concurrence on the same simulated data.
  • Unsupervised KMeans recovers the two physical classes without labels, indicating a natural geometric separation in measurement space rather than an artifact of supervised fitting.
  • Because logistic regression degrades under class imbalance and limited data while nonlinear and ensemble methods do not, the separable-entangled decision boundary is effectively nonlinear in SIC-POVM count space.

Where Pith is reading between the lines

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

  • If the Poisson assumption holds in a real experiment, the same pipeline could be deployed online: a fixed trained regressor would give per-state concurrence estimates in microseconds, enabling high-throughput screening of many two-qubit sources.
  • The concentration of errors at p=1/3 suggests a testable prediction: on real data with known Werner parameter, misclassification rates should peak at the same boundary; observing that would corroborate the learned representation.
  • The trained decision functions act as implicit separability criteria in SIC-POVM count space, potentially applicable as fast experimental tests for states outside the Werner family, though that transfer is untested.
  • Adding detector inefficiency or dark counts would shift the Poisson mean and could break the learned mapping; a natural stress test is to retrain on data generated with losses and see how accuracy degrades.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper explores whether machine learning can classify and quantify entanglement directly from simulated noisy SIC-POVM coincidence-count vectors for two-qubit states, without density-matrix reconstruction. Two datasets are generated: (i) 10,000 separable product states with Bloch radii in {0.4, 0.8, 1.0} versus 10,000 maximally entangled Bell states under local unitaries, and (ii) 250,000 locally rotated Werner states with the Werner parameter discretized in steps of 0.02. Photon counts are generated according to Born probabilities and independent Poisson statistics with N=40,000. The authors report near-perfect classification for separable-versus-Bell states (Tables 1 and 3), high classification accuracy for Werner states (Table 8), and high concurrence-regression accuracy (Table 10), together with a comparison against quantum state tomography. They conclude that ML provides an efficient alternative to standard tomography.

Significance. If the claims were established at the stated level of generality, the paper would be a useful contribution to fast entanglement characterization from raw photon-count data. The generative pipeline is internally consistent and clearly described, including an analytical consistency check for the concurrence of Werner states. However, the tested state families are narrow, all metrics are point estimates from single splits, no code or dataset is shipped, and the QST comparison is confounded by sample size. As a proof of concept for the specific families considered, the results are plausible; as a demonstration of general ML-based entanglement identification, the evidence is incomplete.

major comments (3)
  1. [§3.1, Eq. (A10); §3.2, Eqs. (B10)–(B11)] The broad claim in the abstract and conclusions that ML can identify and quantify entanglement from SIC-POVM data is not supported by the tested state ensembles. In the first dataset, separable states are only product states with Bloch radii in {0.4,0.8,1.0} and entangled states are only maximally entangled Bell states; in the second, the Werner family is a one-parameter set with local-unitary copies. These families are far apart in the 16-dimensional measurement space, so the near-perfect accuracies in Tables 1, 3, and 8 may reflect family-specific separation rather than a general entanglement/separability distinction. The caveat in §4.3 that the study is a proof of concept should be reflected in the abstract and conclusions, or the experiments should be extended to generic two-qubit states (e.g., random states labeled by the Peres–Horodecki criterion).
  2. [§4.2.2, Table 10] The comparison with quantum state tomography is not matched. QST and SVR use 5,000 observations, while the Random Forest Regressor is trained and tested on 250,000 observations. The conclusion that RFR outperforms QST is therefore confounded by sample size. Please run RFR on the same 5,000-observation split as QST/SVR, or run QST on a comparably large sample, and report matched metrics. Also clarify how the least-squares QST reconstruction handles non-physical density matrices with negative eigenvalues.
  3. [§4.1.1, Tables 1 and 3; §4.2, Table 8; §4.2.2, Table 10] All reported metrics are point estimates from a single train/test split, with no error bars, confidence intervals, or repeated-seed experiments. Since the data-generation process is stochastic and some algorithms have random components, differences such as SVC vs RandomForest in Table 8 (0.99498 vs 0.99132) are not shown to be stable. Report means and standard deviations over multiple random splits, or bootstrap confidence intervals, for the main classification and regression results.
minor comments (4)
  1. [§4.3, paragraph 2] The text states that in the 40-example scenario 'all selected algorithms maintain an accuracy of 100%', but Table 3 reports 0.8487 for Logistic Regression and 0.8828 for Decision Tree. This discrepancy should be corrected.
  2. [Data Availability] No code or dataset repository is provided; 'raw data will be made available by the corresponding author on request' is insufficient for reproducibility. Please deposit the generation code and datasets.
  3. [§4.1.1 and §4.3] The inference that the decision boundary is nonlinear because logistic regression degrades with 40 training examples is not justified: with 4,000 training examples, logistic regression achieves accuracy 1.0. The 40-example behavior reflects sample complexity rather than boundary curvature.
  4. [Appendix A, Eq. (A4)] The local unitary is applied only to the second qubit. Please state explicitly whether this generates all maximally entangled two-qubit pure states, and if so, justify it; otherwise, the ensemble of entangled states may be a restricted subset.

Circularity Check

0 steps flagged

No circularity: supervised labels are computed from ground-truth density matrices, and the self-referential simulator benchmark is a generalization caveat, not a reduction.

full rationale

The paper makes no derivational claim that reduces to its inputs. The regression target C(ρ) is computed from each density matrix via the spin-flip construction (Eqs. B13–B17), independently of the 16-dimensional SIC-POVM count vector; the count vector is generated from the same state through Born's rule and Poisson sampling (Eqs. B25–B26), so the supervised task is a legitimate learning of a well-defined functional relation, not a fit renamed as a prediction. Classification labels are likewise assigned from the Werner parameter p (C_bin = 1 iff p > 1/3), not from the measurement vector. The QST baseline is a genuine independent comparison on the same simulated counts. Self-citations ([25], [29]) are bibliographic and not load-bearing; no uniqueness theorem or ansatz is imported from the authors' prior work. The acknowledged limitations—Poisson-only noise, restricted state families, and proof-of-concept status (Sec. 4.3)—are external-validity caveats, not circularity. Training and test data sharing the same simulator affects generalization to real experiments but does not make any claimed prediction equivalent to its input by construction. Therefore no circular step is identified.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central results rest on the simulation formalism and the chosen sampling distributions. The main free parameters are the count scale, the state-ensemble supports, and the ML hyperparameters. No new physical entities are introduced. The domain assumptions about Poisson-only noise and in-distribution training are the most important unverified premises.

free parameters (4)
  • Average detected photon pairs per setting N = 40,000
    Set by hand (Eq. A13/B27); controls the Poisson noise level and therefore how easy classification/regression is. Not fitted to data.
  • Bloch radii for generated separable states = 0.4, 0.8, 1.0
    Chosen sampling support in App. A.2; restricts the separable-state ensemble and keeps the first classification task far from the separability boundary.
  • Werner parameter discretization and unitary grid = Δp=0.02; grid over α, β, γ
    Chosen sampling densities (App. B.2) determine class balance, boundary difficulty, and dataset size; affects all Werner-state results.
  • ML hyperparameters = SVR C=10, ε=0.01; RFR 200 trees; 80/20 split
    Chosen without a tuning study; affects reported accuracy/R². Other classifier hyperparameters (k, kernel details, max depth) are largely scikit-learn defaults.
axioms (5)
  • standard math Born rule: p_k = Tr(M_k ρ) gives the probability of each SIC-POVM outcome.
    Used throughout §3 and Appendices A/B to generate measurement probabilities from known density matrices.
  • standard math The 16 two-qubit SIC-POVM operators M_ij = P_i ⊗ P_j form an informationally complete measurement.
    Invoked in §2.3; ensures the 16-dimensional count vector contains full tomographic information. Cited to Refs. [18,27].
  • domain assumption Photon coincidence counts follow independent Poisson distributions with mean N p_k and no other noise sources.
    Eq. A12/B26. This is the core noise model and the main simplification: losses, dark counts, dead time, and multi-pair emission are excluded.
  • standard math Werner states are separable iff p ≤ 1/3, with concurrence C = max(0, (3p−1)/2).
    Used in App. B.3 (Eqs. B21–B23) to assign labels and targets; follows from the standard Werner-state analysis [30].
  • standard math Local unitary transformations preserve concurrence and entanglement.
    Used to expand the dataset geometry in App. A.1 and B.2 (Eq. B24); standard property of entanglement monotones.

pith-pipeline@v1.3.0-alltime-deepseek · 15545 in / 14084 out tokens · 143663 ms · 2026-08-01T04:21:08.010911+00:00 · methodology

0 comments
read the original abstract

In this work, we investigate the application of Machine Learning (ML) algorithms to the identification and quantitative characterization of quantum entanglement in polarization-entangled photon pairs. The analysis is based on simulated symmetric, informationally complete, positive operator-valued measure (SIC-POVM) measurement data, where each two-qubit state is represented by a 16-dimensional measurement vector corresponding to experimentally accessible coincidence counts. The generated SIC-POVM measurement data include Poissonian shot noise. Several supervised ML algorithms, including Logistic Regression, k-Nearest Neighbors, Decision Trees, Support Vector Machines, and Random Forests, are applied to the classification of separable and entangled states directly from raw measurement data, without explicit density matrix reconstruction or the use of conventional separability criteria. The study additionally explores clustering methods and nonlinear regression techniques for estimating continuous entanglement measures. The obtained results demonstrate that ML methods can achieve very high classification accuracy, even under extremely limited training conditions. These findings indicate that ML may provide an efficient alternative to conventional quantum-state analysis under simulated Poissonian noise conditions.

Figures

Figures reproduced from arXiv: 2607.22853 by Artur Czerwinski, Maciej Wi\'sniewski, Pawe{\l} Moszczy\'nski.

Figure 1
Figure 1. Figure 1: FIG. 1. Dependence of inertia on the number of clusters in the [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Histograms showing the distribution of misclassified states as a function of the Werner-state parameter [PITH_FULL_IMAGE:figures/full_fig_p012_2.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

45 extracted references · 2 linked inside Pith

  1. [1]

    INTRODUCTION Quantum entanglement is one of the most distinctive and nonclassical features of quantum mechanics. Since its original discussion in the celebrated Einstein–Podolsky–Rosen paradox and the subsequent formulation of Bell inequalities, entanglement has evolved from a conceptual curiosity into a fundamental resource for quantum infor- mation scie...

    Pith/arXiv arXiv 2026

  2. [2]

    Density Matrix Representation The mathematical description of quantum systems is based on the formalism of Hilbert spaces and linear opera- tors acting upon them

    QUANTUM STATES AND ENTANGLEMENT FORMALISM 2.1. Density Matrix Representation The mathematical description of quantum systems is based on the formalism of Hilbert spaces and linear opera- tors acting upon them. While isolated quantum systems can be described by state vectors, practical quantum-optical experiments often involve incomplete information, stati...

  3. [3]

    GENERATION OF QUANTUM STATES AND MEASUREMENT DATA The performance of ML algorithms strongly depends on the quality, diversity, and physical realism of the training data. Since large experimentally acquired datasets of entangled photonic states are not readily available, the present study relies on numerically generated quantum states and simulated tomogra...

  4. [4]

    The analysis is organized according to the two datasets considered in this work

    APPLICATION OF ML ALGORITHMS: RESULTS AND ANALYSIS In this section, ML methods are applied to the generated SIC-POVM measurement data. The analysis is organized according to the two datasets considered in this work. In the first part, classification and clustering of separable and maximally entangled states are investigated, representing clearly separated...

  5. [5]

    The analysis was based on pseudo-experimental measurement data generated us- ing SIC-POVM operators

    CONCLUSIONS In this work, we investigated the application of ML methods to the analysis of two-qubit quantum states describing polarization-entangled photon pairs. The analysis was based on pseudo-experimental measurement data generated us- ing SIC-POVM operators. The generated data modelled Poissonian photon-counting noise under controlled simulated cond...

  6. [6]

    Nielsen and Isaac L

    Michael A. Nielsen and Isaac L. Chuang.Quantum Computation and Quantum Information. Cambridge University Press, Cambridge, 2000

    2000

  7. [7]

    Quantum entanglement.Rev

    Ryszard Horodecki, Paweł Horodecki, Michał Horodecki, and Karol Horodecki. Quantum entanglement.Rev. Mod. Phys., 81:865–942, Jun 2009

    2009

  8. [8]

    C. K. Hong and L. Mandel. Theory of parametric frequency down conversion of light.Phys. Rev. A, 31:2409–2418, Apr 1985

    1985

  9. [9]

    Y. H. Shih, A. V. Sergienko, Morton H. Rubin, T. E. Kiess, and C. O. Alley. Two-photon entanglement in type-ii parametric down-conversion.Phys. Rev. A, 50:23–28, Jul 1994

    1994

  10. [10]

    Quantum communication.Nat

    Nicolas Gisin and Rob Thew. Quantum communication.Nat. Photonics, 1(3):165–171, 2007

    2007

  11. [11]

    Andersen, Leonardo Banchi, Mario Berta, Darius Bunandar, Roger Colbeck, Dirk Englund, Tobias Gehring, Cosmo Lupo, Carlo Ottaviani, Joseph Pereira, Mohsen Razavi, Jonatan S

    Stefano Pirandola, Ulrik L. Andersen, Leonardo Banchi, Mario Berta, Darius Bunandar, Roger Colbeck, Dirk Englund, Tobias Gehring, Cosmo Lupo, Carlo Ottaviani, Joseph Pereira, Mohsen Razavi, Jonatan S. Shaari, Marco Tomamichel, Vladyslav C. Usenko, Giuseppe Vallone, Paolo Villoresi, and Petros Wallden. Advances in quantum cryptography.Adv. Opt. Photonics.,...

    2020

  12. [12]

    Separability criterion for density matrices.Phys

    Asher Peres. Separability criterion for density matrices.Phys. Rev. Lett., 77(8):1413–1415, 1996

    1996

  13. [13]

    Separability of mixed states: Necessary and sufficient conditions.Phys

    Michał Horodecki, Paweł Horodecki, and Ryszard Horodecki. Separability of mixed states: Necessary and sufficient conditions.Phys. Lett. A, 223(1–2):1–8, 1996

    1996

  14. [14]

    Entanglement of a pair of quantum bits.Phys

    Sam A Hill and William K Wootters. Entanglement of a pair of quantum bits.Phys. Rev. Lett., 78(26):5022, 1997

    1997

  15. [15]

    Wootters

    William K. Wootters. Entanglement of formation of an arbitrary state of two qubits.Phys. Rev. Lett., 80(10):2245–2248, 1998

    1998

  16. [16]

    Machine learning and the physical sciences.Rev

    Giuseppe Carleo, Ignacio Cirac, Kyle Cranmer, Laurent Daudet, Maria Schuld, Naftali Tishby, Leslie Vogt-Maranto, and Lenka Zdeborová. Machine learning and the physical sciences.Rev. Mod. Phys., 91(4):045002, 2019

    2019

  17. [17]

    Quantum machine learning.Nature, 549(7671):195–202, 2017

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Rebentrost, Nathan Wiebe, and Seth Lloyd. Quantum machine learning.Nature, 549(7671):195–202, 2017

    2017

  18. [18]

    Neural networks for quantum state tomography with constrained measurements.Quantum Inf

    Hailan Ma, Daoyi Dong, Ian R Petersen, Chang-Jiang Huang, and Guo-Yong Xiang. Neural networks for quantum state tomography with constrained measurements.Quantum Inf. Process., 23(9):317, 2024. 16

    2024

  19. [19]

    Entanglement detection with artificial neural networks.Sci

    Naema Asif, Uman Khalid, Awais Khan, Trung Q Duong, and Hyundong Shin. Entanglement detection with artificial neural networks.Sci. Rep., 13(1):1562, 2023

    2023

  20. [20]

    Entanglement detection with classical deep neural networks.Sci

    Julio Ureña, Antonio Sojo, Juani Bermejo-Vega, and Daniel Manzano. Entanglement detection with classical deep neural networks.Sci. Rep., 14(1):18109, 2024

    2024

  21. [21]

    Deep learning of quantum entanglement from incomplete measurements.Sci

    Dominik Koutn` y, Laia Ginés, Magdalena Moczała-Dusanowska, Sven Höfling, Christian Schneider, Ana Predojević, and Miroslav Ježek. Deep learning of quantum entanglement from incomplete measurements.Sci. Adv., 9(29):eadd7131, 2023

    2023

  22. [22]

    Measuring quantum entanglement from local information by machine learning.arXiv preprint arXiv:2209.08501, 2022

    Yulei Huang, Liangyu Che, Chao Wei, Feng Xu, Xinfang Nie, Jun Li, Dawei Lu, and Tao Xin. Measuring quantum entanglement from local information by machine learning.arXiv preprint arXiv:2209.08501, 2022

    Pith/arXiv arXiv 2022

  23. [23]

    Symmetric informationally complete quantum measurements.J

    Joseph M Renes, Robin Blume-Kohout, Andrew J Scott, and Carlton M Caves. Symmetric informationally complete quantum measurements.J. Math. Phys., 45(6):2171–2180, 2004

    2004

  24. [24]

    The sic question: History and state of play.Axioms, 6(3):21, 2017

    Christopher A Fuchs, Michael C Hoang, and Blake C Stacey. The sic question: History and state of play.Axioms, 6(3):21, 2017

    2017

  25. [25]

    Springer Science & Business Media, 2012

    Karl Blum.Density matrix theory and applications, volume 64. Springer Science & Business Media, 2012

    2012

  26. [26]

    Quantum-state estimation.Phys

    Zdenek Hradil. Quantum-state estimation.Phys. Rev. A, 55(3):R1561, 1997

    1997

  27. [27]

    Springer Science & Business Media, 2004

    Matteo Paris and Jaroslav Rehacek.Quantum state estimation, volume 649. Springer Science & Business Media, 2004

    2004

  28. [28]

    Minimal number of operators for observability of n-level quantum systems.Int

    Andrzej Jamiołkowski. Minimal number of operators for observability of n-level quantum systems.Int. J. Theor. Phys., 22(4):369–376, 1983

    1983

  29. [29]

    On complete and incomplete sets of observables, the principle of maximum entropy—revisited

    Andrzej Jamiołkowski. On complete and incomplete sets of observables, the principle of maximum entropy—revisited. Rep. Math. Phys., 46(3):469–482, 2000

    2000

  30. [30]

    Selected concepts of quantum state tomography.Optics, 3(3):268–286, 2022

    Artur Czerwinski. Selected concepts of quantum state tomography.Optics, 3(3):268–286, 2022

    2022

  31. [31]

    Measurement of qubits.Phys

    Daniel FV James, Paul G Kwiat, William J Munro, and Andrew G White. Measurement of qubits.Phys. Rev. A, 64(5):052312, 2001

    2001

  32. [32]

    Minimal qubit tomography.Phys

    Jaroslav Řeháček, Berthold-Georg Englert, and Dagomir Kaszlikowski. Minimal qubit tomography.Phys. Rev. A, 70(5):052321, 2004

    2004

  33. [33]

    Photon, poisson noise

    Samuel W Hasinoff. Photon, poisson noise. InComputer vision: a reference guide, pages 980–982. Springer, 2021

    2021

  34. [34]

    Tomography of time-bin quantum states using time-resolved detection.Phys

    Karolina Sedziak-Kacprowicz, Artur Czerwinski, and Piotr Kolenderski. Tomography of time-bin quantum states using time-resolved detection.Phys. Rev. A, 102(5):052420, 2020

    2020

  35. [35]

    Reinhard F. Werner. Quantum states with einstein-podolsky-rosen correlations admitting a hidden-variable model.Phys. Rev. A, 40(8):4277–4281, 1989

    1989

  36. [36]

    Bishop.Pattern Recognition and Machine Learning

    Christopher M. Bishop.Pattern Recognition and Machine Learning. Springer, 2006

    2006

  37. [37]

    Scikit-learn: Machine learning in python.J

    Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blon- del, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and Édouard Duchesnay. Scikit-learn: Machine learning in python.J. Mach. Learn. Res., 12:2825–2830...

    2011

  38. [38]

    Generation of Maximally Entangled States The entangled-state ensemble was generated from the Bell states |Ψ−⟩= 1√ 2 (|01⟩ − |10⟩),(A1) and |Φ+⟩= 1√ 2 (|00⟩+|11⟩).(A2) To obtain a geometrically diverse collection of maximally entangled states, local unitary transformations were applied according to U(α, β, γ) =e−iασz/2e−iβσy/2e−iγσz/2,(A3) whereσ x,σ y, an...

  39. [39]

    Generation of Separable States The separable-state ensemble was constructed from tensor products of single-qubit density matrices. Each qubit was parametrized using the Bloch-sphere representation rx =rsinθcosϕ,(A5) ry =rsinθsinϕ,(A6) rz =rcosθ.(A7) The corresponding density matrix is given by ρ(r, θ, ϕ) =1 2 (I2 +r xσx +r yσy +r zσz).(A8) Two-qubit separ...

  40. [40]

    Consequently, each quantum state was represented by a sixteen-dimensional measurement vector x= (n 1, n2,

    Generation of Noisy Measurement Data For each density matrixρ, the probability associated with thek-th SIC-POVM outcome was calculated according to Born’s rule, pk = Tr(Mkρ).(A11) The measurement outcomes were generated with Poissonian shot noise, nk ∼Poisson(N pk),(A12) where N= 40,000(A13) denotes the average number of detected photon pairs per measurem...

  41. [41]

    For this purpose, we employed the family of Werner states, which provides a convenient interpolation between maximally mixed and maximally entangled two-qubit states

    Werner States as a Source of Variable Entanglement The regression task considered in this work requires quantum states exhibiting continuously varying degrees of entanglement. For this purpose, we employed the family of Werner states, which provides a convenient interpolation between maximally mixed and maximally entangled two-qubit states. The constructi...

  42. [42]

    However, using only the canonical Werner form would restrict the generated states to a highly symmetric subset of the full two-qubit state space

    Local Unitary Transformations The Werner parameterpcontrols the amount of entanglement. However, using only the canonical Werner form would restrict the generated states to a highly symmetric subset of the full two-qubit state space. To increase the geometrical diversity of the dataset, local unitary transformations were applied independently to both qubi...

  43. [43]

    Concurrence for Two-Qubit States For each generated quantum state, the amount of entanglement was quantified using concurrence. For an arbitrary two-qubit density matrixρ, the spin-flipped state is defined as eρ= (σy ⊗σ y)ρ∗(σy ⊗σ y),(B13) whereρ ∗ denotes complex conjugation in the computational basis, and σy = 0−i i0 (B14) is the Pauli-Ymatrix. Next, th...

  44. [44]

    For thek-th measurement operatorM k, the corresponding probability was evaluated using Born’s rule: pk = Tr Mkρ(LU) W , k= 1,

    Generation of Noisy SIC-POVM Measurement Data For each locally transformed Werner state, the SIC-POVM measurement operators introduced in Section 2.3 were used to generate pseudo-experimental measurement data. For thek-th measurement operatorM k, the corresponding probability was evaluated using Born’s rule: pk = Tr Mkρ(LU) W , k= 1, . . . ,16.(B25) To em...

  45. [45]

    , n16, C,Cbin, p .(B28) Equivalently, the ML input vector was x= (n 1, n2,

    Construction of the Regression Dataset For each generated quantum state, the final data record consisted of sixteen noisy SIC-POVM counts and the corresponding concurrence value: ρ(LU) W ← → n1, n2, . . . , n16, C,Cbin, p .(B28) Equivalently, the ML input vector was x= (n 1, n2, . . . , n16),(B29) where the regression target wasy=Cand for classification t...