Pith. sign in

REVIEW 3 major objections 7 minor 20 references

A passive optical tap can identify which quantum communication protocol is running, with up to 96% accuracy, without collapsing entanglement.

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-4.5

2026-07-31 10:17 UTC pith:G6B7NKIV

load-bearing objection Solid bench demo that four protocol implementations leave passive side-channel fingerprints, but the 96% figure mostly tracks lab-specific HWP timing, not abstract protocol semantics. the 3 major comments →

arxiv 2607.24624 v1 pith:G6B7NKIV submitted 2026-07-27 quant-ph cs.NI

Experimental Protocol Fingerprinting in Quantum Networks via Physical Layer Side Channel Analysis

Lance Young , Contessa Wilburn , Carrie Houston , Blaine Keyton , Marwan Elawady , Mohamed Shaban , Muhammad Ismail This is my paper
classification quant-ph cs.NI
keywords protocol fingerprintingphysical layer side channelquantum communicationoptical tappingentanglement distributionheralded QKDquantum identity authenticationphoton detection statistics
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.

This paper asks whether quantum networks leak protocol identity the way classical systems leak secrets through side channels. On a polarization-entangled photon link, the authors run four protocols—entanglement distribution, quantum gate sequences, heralded QKD, and quantum identity authentication—and let an observer see only a fraction of the light via a beam splitter, never measuring the encoded quantum states. From photon arrival timing and optical power alone, a sequence model can name the active protocol with accuracy up to 96% when 30% of the light is tapped, and still 70–89% when only 10% is taken. Bell tests stay above the classical bound, so the tap does not destroy the entanglement the network relies on. The result matters because protocol identity is often the first thing an adversary needs before targeting a service, and the paper shows that identity can be read without the usual quantum alarms.

Core claim

Different quantum communication protocols leave distinguishable physical-layer signatures in photon detection statistics and optical power. Under a passive tap that never measures the encoded states, those signatures are enough to classify four representative protocols on a polarization-entangled link at up to 96% accuracy (30:70 split) and 70–89% (10:90 split), while CHSH S-values remain above 2, confirming entanglement is preserved.

What carries the argument

Passive optical tapping plus a data-driven fingerprint: a beam splitter samples a fixed fraction of one photon path; time-tagger features (count rate, interarrival mean/std/CV) and power-meter features (irradiance and power statistics) feed sequence models (best: Bi-Stacked LSTM), with SHAP showing timing features dominate and power features add complementary signal.

Load-bearing premise

The measured differences come from how the protocols behave, not from the particular lab source, wave-plate schedules, heralding setup, or low-noise free-space conditions used in the experiment.

What would settle it

Repeat the same four protocols on an independent entangled-photon link (different source, fiber or free-space path, and detector set) with a 10:90 or 30:70 tap; if classification accuracy collapses to chance while S-values still exceed 2, the claimed protocol-level side channel does not generalize.

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

If this is right

  • Protocol identity in quantum networks can be treated as a side-channel secret that must be protected, not assumed private by quantum mechanics alone.
  • Even a 10% optical tap can support traffic analysis and usage profiling without breaking entanglement or triggering state-collapse alarms.
  • Non-intrusive network monitors could use the same tap-and-classify pipeline for diagnostics, not only adversarial ends.
  • Defenses must now target temporal and intensity signatures (obfuscation, rate padding, or signature equalization), not only quantum-state secrecy.
  • Higher tap fractions buy accuracy at a clear cost in photon and coincidence rate, defining an observability-versus-rate tradeoff operators must manage.

Where Pith is reading between the lines

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

  • If timing features dominate, simple rate-smoothing or deliberate burst shaping at the source may be a cheaper countermeasure than optical isolation of the channel.
  • The same fingerprinting approach likely extends to other SPDC-based services (teleportation, entanglement swapping) whose duty cycles and basis choices also imprint on arrival statistics.
  • A realistic adversary who can only afford a few-percent tap may still succeed by aggregating longer observation windows, so future work should report accuracy versus both split ratio and integration time.
  • Standardization bodies for quantum networks may need side-channel evaluation criteria analogous to those already used for classical cryptographic modules.

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 / 7 minor

Summary. The manuscript introduces the problem of "protocol fingerprinting" in quantum networks: can a passive observer who taps a fraction of the optical signal (without measuring the encoded quantum states) infer which quantum communication protocol is running? Using a polarization-entangled SPDC testbed, the authors implement four protocols (entanglement distribution, quantum gate sequences, heralded QKD, and QIA), collect photon-detection timing statistics and optical power measurements at 10:90 and 30:70 tap ratios, and train recurrent sequence models (best: Bi-Stacked LSTM) to classify protocol identity. Reported accuracies reach 96% at 30:70 and 70–89% at 10:90. CHSH measurements (S = 2.324–2.359, all above 2) are used to argue the tap preserves entanglement, and SHAP analysis shows timing features (count rate, interarrival statistics) dominate classification. The problem formulation is genuinely novel and the experimental execution is competent; however, two methodological issues bear directly on what the results demonstrate, and the security interpretation is currently stated more strongly than the evidence supports.

Significance. If the central claim holds, the work identifies a real and previously underexplored leakage channel: protocol-level traffic analysis on quantum links by a passive adversary, analogous to classical website fingerprinting. Strengths worth naming: (i) the problem is new — prior quantum side-channel work targets QKD internals or quantum-computer controllers, not cross-protocol identification on a communication link; (ii) the same-hardware multi-protocol testbed is a reasonable design for isolating protocol-induced differences; (iii) the entanglement-preservation check under tapping (CHSH with uncertainties) is a genuine experimental contribution rather than an assumption; (iv) the inclusion of SHAP-based interpretability and confusion-matrix error analysis is good practice and lets the reader see *which* physical features carry the signal. The manuscript is also honest about the observability/rate tradeoff. What limits significance is that the demonstrated separability may be a property of these four bench implementations (hand-set motorized waveplate schedules, protocol-specific herald roles) rather than of the protocols' logical content — the paper's own SHAP result points this way —

major comments (3)
  1. [§II-C and §III-A (vs. §V-F)] The load-bearing interpretive claim is that observable differences 'arise from protocol behavior rather than changes in the underlying physical infrastructure' (§II-C, final sentence). Shared hardware does not establish this. Per §III-A, each protocol is assigned a distinct, hand-set temporal pattern on the same apparatus: gate sequences use fixed HWP schedules ({0°,45°,0°,45°} or {0°,22.5°,45°,67.5°}), entanglement distribution uses no modulation, QKD uses random four-state preparation, and QIA uses key-dependent interleaved authentication rounds. The paper's own SHAP analysis (§V-F, Figs. 4a/4c) shows photon count rate and interarrival-time statistics dominate — features that directly encode the cadence of the motorized waveplate schedule and herald usage, not the protocols' logical content. In a deployed link, BB84-style QKD and entanglement distribution would share near-identical fas
  2. [§III-D / §IV-A] The evaluation protocol permits run-level leakage. Data are collected 'over multiple runs' per protocol, segmented, randomly arranged, and split 80/20 (7200 samples → 1440 sequences → 1152/288). Because protocols were necessarily executed in separate runs (the HWP schedule and herald configuration differ per protocol), run-specific calibration — alignment drift, count-rate offsets, temperature — is perfectly correlated with the label, and a random segment-level split lets the model exploit it. Local min–max normalization (§III-D) mitigates but does not eliminate this, since normalization parameters are themselves run-level statistics. The reported accuracies (Table I) are therefore an upper bound on protocol separability. A held-out-run (or held-out-day) evaluation — train on all runs but one per class, test on the excluded run — is the standard control and is feasible with the existing
  3. [§V-A and §V-G] The 'non-destructive observation' framing needs qualification. CHSH S > 2 under tapping shows that the *surviving* pairs remain entangled; it does not show the tap is undetectable. The 30:70 configuration — the one yielding the headline 96% accuracy — reduces photon count and coincidence rates by ~29% each (Fig. 2), a macroscopic, easily monitored signature that any reasonably instrumented link would flag. Only the 10:90 configuration (~6%/2% drops) is plausibly stealthy, and there the accuracy is 70–89%. The abstract and §V-G should state this tradeoff as a constraint on the threat model rather than describing the observation model as non-destructive without qualification. Relatedly, the threat model (§II-B) assumes 'no prior knowledge of protocol family or type,' yet the classifier is trained on labeled traces of exactly these four protocols — a closed-world assumption that should be s
minor comments (7)
  1. [§V-E] The sentence 'Overall, the confusion matrix analysis reveals that classification errors are not random...' and the following sentence are duplicated nearly verbatim in consecutive paragraphs.
  2. [§I-B] 'Shapely additive explanations (SHAP)' should read 'SHapley Additive exPlanations'.
  3. [References] Ref. [15], cited for SHAP in §I-B and §IV-B, is a secondary arXiv preprint on feature selection; the primary citation (Lundberg & Lee, already listed as [20]) should be used at first mention.
  4. [§III-D / Table I] With only 288 test sequences, the headline 96.18% corresponds to 277/288 correct; binomial 95% CI is roughly ±2.2%. Confidence intervals (or per-run variance across repetitions) should accompany Table I, especially given the single train/test split.
  5. [Abstract] Typesetting: missing spaces around numbers ('up to96%under30:70...'); also 'up to 96%' is achieved by one model under one configuration — the abstract should indicate the range across models to avoid overreading.
  6. [§III-A] The statement that θ = 45° implements a Pauli-X and θ = 22.5° a Hadamard refers to HWP angle conventions acting on polarization; a half-sentence clarifying that these are polarization rotations (HWP at θ rotates linear polarization by 2θ) would prevent confusion with qubit-gate angles.
  7. [§II-C, Fig. 1] The tap is placed after the 808 nm HWP; it would help to state explicitly whether the observer's detector is polarization-insensitive (no analyzer before the observer detector), since this determines whether state-preparation information is directly visible at the tap or only via rate/timing.

Circularity Check

0 steps flagged

No circular derivation: empirical protocol classification on held-out physical-layer measurements.

full rationale

This paper is an experimental side-channel classification study, not a first-principles derivation. Protocol labels are experimenter-chosen configurations (entanglement distribution, gate sequences, heralded QKD, QIA) run on a shared SPDC testbed; features are measured photon timestamps and optical power; models (RNN/LSTM/GRU variants) are trained and scored on an 80/20 held-out split of sequence windows. Reported accuracies (up to ~96% at 30:70 tap, ~70–89% at 10:90) and CHSH S-values are empirical outcomes, not quantities forced by fitting a parameter that is then renamed as a prediction. The only author-overlapping citation of note is [19] (Shaban & Ismail), which merely specifies the QIA protocol encoding they implement; it is not used as a uniqueness theorem, ansatz, or load-bearing premise for the fingerprinting claim. No equation equates a claimed prediction to its fitted input by construction, and no self-citation chain substitutes for the experimental result. Methodological concerns about whether distinguishability reflects protocol semantics versus lab-specific HWP schedules or run structure are validity/generalization issues, not circularity. Score 0 is therefore appropriate.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 1 invented entities

The central claim rests on standard quantum-optics and ML practice plus experimental design choices, not on new physical entities. Load-bearing inputs are the passive-tap threat model, the four protocol instantiations on one SPDC link, chosen tap ratios, engineered timing/power features, and trained RNN classifiers. No graviton-like invention; free parameters are experimental and ML knobs that affect reported accuracy.

free parameters (5)
  • optical tap ratios (10:90 and 30:70) = 10:90 and 30:70
    Discrete observation strengths chosen by the experimenters; primary accuracy claims are reported at these two points only.
  • sequence length (5 consecutive samples) and dataset construction (7200→1440 sequences, 80/20 split) = 5 samples/sequence; 1152 train / 288 test
    Temporal windowing and split choices directly shape the supervised learning task and reported metrics.
  • Bi-Stacked LSTM and related RNN hyperparameters = 64/32 units, dropout 0.3, lr 1e-4
    Hidden sizes 64/32, dropout 0.3, Adam lr 1e-4, dense 32 tanh—tuned empirically for best accuracy; best-model claims depend on this fit.
  • gate-sequence HWP angle schedules = {0°,45°,0°,45°} and {0°,22.5°,45°,67.5°}
    Two specific polarization sequences define the ‘quantum gate sequences’ class; distinguishability may depend on these hand-chosen patterns.
  • percentile-based outlier clipping thresholds
    Preprocessing clips used to suppress detector/background artifacts; thresholds not numerically specified but affect features.
axioms (5)
  • domain assumption SPDC in crossed BBO with compensation produces a usable polarization Bell state |ψ⟩∝|HH⟩+e^{iφ}|VV⟩ verifiable by CHSH.
    Standard entangled-photon source assumption; §II-C and entanglement validation in §V-A.
  • domain assumption A beam-splitter tap after state preparation yields only aggregate photon statistics/power and does not constitute a direct quantum-state measurement that collapses the encoded information in the threat model’s sense.
    Core passive-observer model in §II-B; underwrites ‘non-destructive side channel’ language.
  • domain assumption Differences among the four implemented protocols dominate physical-layer signatures relative to shared hardware drift when runs are normalized per experiment.
    Stated motivation for shared testbed and local min-max normalization (§II-C, §III-D).
  • ad hoc to paper Supervised sequence models trained on labeled lab traces generalize to the held-out segments of the same campaign (i.i.d. enough after random segment arrangement).
    Standard ML assumption; no cross-day or cross-setup validation reported (§III-D, §IV-A).
  • domain assumption CHSH S>2 is sufficient experimental evidence that entanglement relevant to the protocols remains intact under tapping.
    Used in abstract and §V-A to validate non-destructive observation.
invented entities (1)
  • protocol fingerprint (physical-layer feature signature of a quantum communication protocol) independent evidence
    purpose: Name the observable timing/power pattern used as the classification target for protocol identity.
    Operational/empirical construct from measured features, not a new physical substance; no independent ontology beyond the ML task.

pith-pipeline@v1.2.0-grok45-kimik3 · 16483 in / 3777 out tokens · 72908 ms · 2026-07-31T10:17:51.308366+00:00 · methodology

0 comments
read the original abstract

Quantum communication is a key enabler of next-generation networks, leveraging quantum entanglement to enable a new class of information exchange. While prior work has focused on the theoretical analysis of communication protocols, their exposure to physical layer side channel analysis remains largely unexplored. In classical systems, side channel analysis has been shown to reveal sensitive information without accessing the underlying data, raising the question of whether similar risks exist in quantum networks. In this work, we investigate whether different quantum communication protocols exhibit distinguishable signatures that can be inferred through passive side channel observations. We consider a threat model in which an observer accesses only a fraction of the optical signal without directly measuring the encoded quantum states. Under this setting, we experimentally examine four representative protocols, namely entanglement distribution, quantum gate sequences, heralded quantum key distribution, and quantum identity authentication, realized on a polarization entangled photon link. Observable physical layer features, including single photon detection statistics and optical power measurements, are collected and used to construct protocol fingerprints. We develop a data-driven framework for protocol identification based on these observations. Our results show that protocol identity can be inferred with accuracy reaching up to 96% under 30:70 sampling configuration/optical tapping, while remaining distinguishable at 10:90 with accuracy ranging from 70-89%. Bell inequality measurements confirm that the sampling/tapping process preserves entanglement, validating the non-destructive nature of the observation model. These findings demonstrate that side channel analysis can expose protocol-level information without disrupting quantum correlations, introducing new security considerations.

Figures

Figures reproduced from arXiv: 2607.24624 by Blaine Keyton, Carrie Houston, Contessa Wilburn, Lance Young, Marwan Elawady, Mohamed Shaban, Muhammad Ismail.

Figure 1
Figure 1. Figure 1: Quantum communication testbed used for protocol fingerprinting via physical layer observations. The setup includes (1) [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of passive sampling on photon count and coinci [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Confusion matrices illustrating the impact of sampling ratio and feature type on protocol classification. (a) Time tagger [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: SHAP feature importance analysis for different sampling configurations and feature sets: (a) [PITH_FULL_IMAGE:figures/full_fig_p009_4.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

20 extracted references · 2 canonical work pages

  1. [1]

    Security vulnerabilities in quantum cloud systems: A survey on emerging threats,

    J. Coupel and T. Farheen, “Security vulnerabilities in quantum cloud systems: A survey on emerging threats,” 2025. [Online]. Available: https://arxiv.org/abs/2504.19064

    Pith/arXiv arXiv 2025

  2. [2]

    Fingerprinting (columbia university cs6772 paper),

    D. Rubensteinet al., “Fingerprinting (columbia university cs6772 paper),” inCourse Papers, Columbia University, 2006. [Online]. Available: https://www.cs.columbia.edu/ ∼danr/6772/papers/fingerprint. pdf

    2006

  3. [3]

    Special issue on “side channel attacks

    S. Hong, “Special issue on “side channel attacks”,”Applied Sciences, vol. 9, no. 9, 2019. [Online]. Available: https://www.mdpi.com/ 2076-3417/9/9/1881

    2019

  4. [4]

    Spectral side channels in quantum key distribution under laser damage,

    B. Gao, J. Liu, E. Borisova, H. Tan, M. Zhong, Z. Chen, Q. Peng, W. Shi, A. Ponosova, V . Makarov, and A. Huang, “Spectral side channels in quantum key distribution under laser damage,” 2025. [Online]. Available: https://arxiv.org/abs/2512.11701

    arXiv 2025

  5. [5]

    Estimating distinguishability measures on quantum computers,

    S. Rethinasamy, R. Agarwal, K. Sharma, and M. M. Wilde, “Estimating distinguishability measures on quantum computers,” Physical Review A, vol. 108, no. 1, Jul. 2023. [Online]. Available: http://dx.doi.org/10.1103/PhysRevA.108.012409

    doi:10.1103/physreva.108.012409 2023

  6. [6]

    Electromagnetic side-channel attack risk assessment on a practical quantum-key- distribution receiver based on multi-class classification,

    J. J. Pantoja, V . A. Bucheli, and R. Donaldson, “Electromagnetic side-channel attack risk assessment on a practical quantum-key- distribution receiver based on multi-class classification,”EPJ Quantum Technology, vol. 11, no. 1, p. 78, 2024. [Online]. Available: https://doi.org/10.1140/epjqt/s40507-024-00290-6

    work page doi:10.1140/epjqt/s40507-024-00290-6 2024

  7. [7]

    Side channel vulnerability in parity computation of generic key reconciliation process on QKD,

    G. Kim, D. Park, H. Kim, and S. Hong, “Side channel vulnerability in parity computation of generic key reconciliation process on QKD,” in2021 International Conference on Information and Communication Technology Convergence (ICTC), 2021, pp. 257–261

    2021

  8. [8]

    Single trace side channel analysis on quantum key distribution,

    S. Kim, S. Jin, Y . Lee, B. Park, H. Kim, and S. Hong, “Single trace side channel analysis on quantum key distribution,” in2018 International Conference on Information and Communication Technology Conver- gence (ICTC), 2018, pp. 736–739

    2018

  9. [9]

    Experimental side channel analysis of BB84 QKD source,

    A. Biswas, A. Banerji, P. Chandravanshi, R. Kumar, and R. P. Singh, “Experimental side channel analysis of BB84 QKD source,”IEEE Journal of Quantum Electronics, vol. 57, no. 6, pp. 1–7, 2021

    2021

  10. [10]

    Deep-learning-based radio-frequency side-channel attack on quantum key distribution,

    A. Baliuka, M. St ¨ocker, M. Auer, P. Freiwang, H. Weinfurter, and L. Knips, “Deep-learning-based radio-frequency side-channel attack on quantum key distribution,”Phys. Rev. Appl., vol. 20, p. 054040, Nov 2023. [Online]. Available: https://link.aps.org/doi/10.1103/ PhysRevApplied.20.054040

    2023

  11. [11]

    Quantum circuit reconstruction from power side-channel attacks on quantum computer controllers,

    F. Erata, C. Xu, R. Piskac, and J. Szefer, “Quantum circuit reconstruction from power side-channel attacks on quantum computer controllers,”IACR Transactions on Cryptographic Hardware and Embedded Systems, vol. 2024, no. 2, p. 735–768, Mar. 2024. [Online]. Available: http://dx.doi.org/10.46586/tches.v2024.i2.735-768

    doi:10.46586/tches.v2024.i2.735-768 2024

  12. [12]

    Quantum leak: Timing side-channel attacks on cloud-based quantum services,

    C. Lu, E. Telang, A. Aysu, and K. Basu, “Quantum leak: Timing side-channel attacks on cloud-based quantum services,” inProceedings of the Great Lakes Symposium on VLSI 2025, ser. GLSVLSI ’25. New York, NY , USA: Association for Computing Machinery, 2025, p. 252–257. [Online]. Available: https://doi.org/10.1145/3716368.3735264

    arXiv 2025

  13. [13]

    Qtime: A machine learning framework for timing side-channel analysis in quantum circuit simulators,

    B. Dong, H. Feng, and Q. Wang, “Qtime: A machine learning framework for timing side-channel analysis in quantum circuit simulators,” in2025 IEEE 43rd International Conference on Computer Design (ICCD), 2025, pp. 335–341

    2025

  14. [14]

    Exploration of power side-channel vulnerabilities in quantum computer controllers,

    C. Xu, F. Erata, and J. Szefer, “Exploration of power side-channel vulnerabilities in quantum computer controllers,” inProceedings of the 2023 ACM SIGSAC Conference on Computer and Communications Security, ser. CCS ’23. New York, NY , USA: Association for Computing Machinery, 2023, p. 579–593. [Online]. Available: https: //doi.org/10.1145/3576915.3623118

    arXiv 2023

  15. [15]

    Shap-select: Lightweight feature selection using shap values and regression,

    E. Kraev, B. Koseoglu, L. Traverso, and M. Topiwalla, “Shap-select: Lightweight feature selection using shap values and regression,” 2024. [Online]. Available: https://arxiv.org/abs/2410.06815

    Pith/arXiv arXiv 2024

  16. [16]

    Ultrabright source of polarization-entangled photons,

    P. G. Kwiat, E. Waks, A. G. White, I. Appelbaum, and P. H. Eberhard, “Ultrabright source of polarization-entangled photons,”Phys. Rev. A, vol. 60, pp. R773–R776, Aug 1999. [Online]. Available: https://link.aps.org/doi/10.1103/PhysRevA.60.R773

    doi:10.1103/physreva.60.r773 1999

  17. [17]

    Proof-of-principle experimental demonstration of quantum gate verification,

    M. Luo, X. Zhang, and X. Zhou, “Proof-of-principle experimental demonstration of quantum gate verification,”Phys. Rev. A, vol. 105, p. 012614, Jan 2022. [Online]. Available: https://link.aps.org/doi/10.1103/ PhysRevA.105.012614

    2022

  18. [18]

    Heralded single-photon sources for quantum-key-distribution applications,

    M. Schiavon, G. Vallone, F. Ticozzi, and P. Villoresi, “Heralded single-photon sources for quantum-key-distribution applications,”Phys. Rev. A, vol. 93, p. 012331, Jan 2016. [Online]. Available: https: //link.aps.org/doi/10.1103/PhysRevA.93.012331

    work page doi:10.1103/physreva.93.012331 2016

  19. [19]

    Secured quantum identity authentication pro- tocol for quantum networks,

    M. Shaban and M. Ismail, “Secured quantum identity authentication pro- tocol for quantum networks,” in2024 IEEE 100th Vehicular Technology Conference (VTC2024-Fall), 2024, pp. 1–6

    2024

  20. [20]

    A unified approach to interpreting model predictions,

    S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” inProceedings of the 31st International Conference on Neural Information Processing Systems, ser. NIPS’17. Red Hook, NY , USA: Curran Associates Inc., 2017, p. 4768–4777

    2017