Entanglement entropy is an attractive candidate for explaining quantum machine-learning models: it is intrinsic to the encoding, cheap to compute, and requires no measurement. We test whether it is a faithful feature attribution for a financial-transaction quantum feature map, under a protocol predeclared down to its thresholds, its comparison family, and its posture on failure. Two questions are separated by construction. First, does the operating point selected for accuracy leave usable attribution signal? Second, does that signal correspond to what actually drives the model’s decisions? On the Kaggle Credit-Card Fraud dataset over ten repeated case-control splits, per-feature entropy dispersion is 71% lower at the accuracy-selected entanglement strength than at the midpoint of the grid, and the difference is significant after Holm–Bonferroni correction; however, a lower-strength regime with 3.4× the dispersion is available at no statistically detectable accuracy cost. Faithfulness fails decisively. Entropy rankings are significantly anti-correlated with model-aware occlusion attributions (median Spearman −0.299), and on a synthetic task whose informative features are known by construction and which the quantum model demonstrably learns, occlusion recovers ground truth (+0.237) while entropy does not (−0.025), both significant after correction. A model-randomisation check identifies the cause exactly: per-feature entropy is invariant to the trained model, with maximum change of exactly zero across all ten seeds, so it cannot distinguish a correctly trained model from one fitted to shuffled labels. We conclude that entanglement entropy is a genuine and inexpensive diagnostic of the encoding but not an explanation of the model, and we report this refutation as predeclared. Balanced-cohort metrics are benchmark values, not deployment-prevalence estimates.
Entanglement entropy is an attractive candidate for explaining quantum machine-learning models: it is intrinsic to the encoding, cheap to compute, and requires no measurement. We test whether it is a faithful feature attribution for a financial-transaction quantum feature map, under a protocol predeclared down to its thresholds, its comparison family, and its posture on failure. Two questions are separated by construction. First, does the operating point selected for accuracy leave usable attribution signal? Second, does that signal correspond to what actually drives the model’s decisions? On the Kaggle Credit-Card Fraud dataset over ten repeated case-control splits, per-feature entropy dispersion is 71% lower at the accuracy-selected entanglement strength than at the midpoint of the grid, and the difference is significant after Holm–Bonferroni correction; however, a lower-strength regime with 3.4× the dispersion is available at no statistically detectable accuracy cost. Faithfulness fails decisively. Entropy rankings are significantly anti-correlated with model-aware occlusion attributions (median Spearman −0.299), and on a synthetic task whose informative features are known by construction and which the quantum model demonstrably learns, occlusion recovers ground truth (+0.237) while entropy does not (−0.025), both significant after correction. A model-randomisation check identifies the cause exactly: per-feature entropy is invariant to the trained model, with maximum change of exactly zero across all ten seeds, so it cannot distinguish a correctly trained model from one fitted to shuffled labels. We conclude that entanglement entropy is a genuine and inexpensive diagnostic of the encoding but not an explanation of the model, and we report this refutation as predeclared. Balanced-cohort metrics are benchmark values, not deployment-prevalence estimates.
Claims about quantum machine learning are routinely established on a single dataset and generalised informally. This paper tests one such claim directly. Earlier work in this series found, under a leakage-controlled predeclared protocol, that a financial-transaction-specific quantum feature map does not outperform matched classical baselines on the Kaggle Credit-Card Fraud dataset. We freeze source-column contracts and an experimental protocol in advance, inherit every predicted direction from that published result, and replicate on CICIDS-2017 (2.8 million flows) and UNSW-NB15 (257,673 records) — six paired comparisons under one Holm–Bonferroni correction. The replication is partial, and the failure is informative. The map’s advantage over naive amplitude encoding replicates on both datasets and is now established across three datasets and two domains. But the prediction that a matched RBF-SVM would win is significantly reversed on UNSW-NB15, where the quantum kernel wins by a median 0.021 F1 — identifying a regime in which the approach pays. That result is bounded: a random forest on identical cohorts outperforms the quantum kernel on both datasets, so no general claim of quantum advantage is supportable. Two further findings weaken the premise of domain-specific encoding itself. The advantage over generic quantum encodings differs in sign between the two network datasets, and the unanimous entanglement strength selected on financial data does not reproduce on either. We conclude that quantum kernel results reported on one dataset should not be assumed to transfer, in either direction, and that predeclared cross-domain replication is cheap enough to be routine. Balanced-cohort metrics are benchmark values, not deployment-prevalence estimates.
The financial-transaction feature map FT-FM, introduced in the first paper of this series, is empirically more expressive than generic quantum encodings but operates in a near-maximally entangled regime in which quantum-kernel values concentrate and kernel-based classification fails. This paper introduces FT-FM-lite, a refinement of FT-FM with a single entanglement-strength hyperparameter β ∈ [0, 1] that preserves the domain-informed encoding axioms while allowing the feature map to operate in the intermediate expressibility–trainability regime; β = 1 recovers FT-FM exactly. A validation-based protocol selects β per training split. In a ten-seed synthetic diagnostic against exactly implemented matched controls, validation-selected FT-FM-lite attains a mean held-out F1 of 0.620 (95% bootstrap CI [0.530, 0.709]), exceeding exact amplitude encoding (0.362) and Qiskit’s ZZFeatureMap (0.033) with Holm–Bonferroni-corrected significance, while its advantage over full-strength FT-FM (0.490) is not significant after correction. We further specify a leakage-controlled, case-control benchmarking protocol with paired Wilcoxon significance testing and apply it to the Kaggle Credit-Card Fraud dataset over ten repeated stratified case-control splits. In contrast to the synthetic diagnostic, validation selection chose full strength (β = 1) in every seed, and FT-FM-lite attained a mean case-control test F1 of 0.891 (95% bootstrap CI [0.870, 0.907]; ROC-AUC 0.971; PR-AUC 0.976) — significantly above exact amplitude encoding and significantly below a matched RBF-SVM after Holm–Bonferroni correction, and statistically indistinguishable from Qiskit’s ZZFeatureMap and the remaining classical baselines. Balanced-cohort metrics are benchmark values, not deployment-prevalence estimates.
The paper is an elaborate comparative assessment of the most relevant entropy-based quantities to be applied in quantifying quantum entanglement in pure, mixed, bipartite, and multipartite quantum systems. The study fills the existing gap in the literature in which most of the previous literature examines individual measures, but a lot of knowledge is lacking on what these measures are regarding their strengths, weaknesses, and the operational usefulness against each other. The study can prove the behaviour of each measure under noise, decoherence, dimensional scaling, and variation of state-type by analysing von Neumann entropy, the Tsallis entropy, entropy of entanglement of formation, concurrence and logarithmic negativity. Numerical studies with pure two-qubit states, Werner states, GHZ states, W states and mixed random states demonstrate specific sensitivity properties and computational limitations which ensure that no single measure can be a universal quantifier. The paper goes on to say that entropy-only measures do not work in mixed regimes because of a dominance in classical correlations. The results have several important implications to quantum communication, quantum simulation, and quantum computing in the NISQ-era, where the choice of measurement has a direct impact on benchmarking, error-correction scheme, and protocol performance based on entanglement. The paper ends with a recommendation of contextual, purpose-based scheme of selecting entanglement measures and suggests a further study on scalable benchmarking, hybrid quantifiers, and operational performance-based entanglement measurements. These observations reinforce the theoretical premise of entanglement studies and enable the advancement of more effective quantum technologies in the field of practice.
Hybrid quantum–classical detectors are usually evaluated as monolithic alternatives to classical models, which conflates two separable questions: whether the hard cases can be identified, and whether a quantum model can resolve them. We introduce Q-RAD, a gated residual architecture that separates these questions by construction. A quantum-blind, quantum-cost-free gating function routes only ambiguous transactions to a financial-transaction-specific quantum-kernel branch, whose scores are fused residually with a classical trunk under an explicit finite-shot budget. We predeclare the architecture, an eight-comparison confirmatory family with Holm–Bonferroni correction, a cost metric in quantum circuit executions, and a refutation posture, then evaluate on the Kaggle Credit-Card Fraud dataset over ten repeated case-control splits under three blocking validity gates, including exact-equality replay of previously published results. The routing half of the architecture succeeds: the gate attains 0.889 AUROC against realized classical errors and captures 77% of them within a 22% routing budget. The quantum half fails: handed exactly those hard cases, the quantum residual produced zero net corrections in nine of ten seeds and one net regression in the tenth. Q-RAD attains mean case-control test F1 0.916 against 0.918 for the classical trunk and 0.922 for an otherwise identical classical second opinion; the hypotheses of improvement over the classical trunk, over random routing, and over a classical second opinion are all refuted after correction. We report this refutation as predeclared, and localise it: ambiguity in this task is identifiable from classical signals alone, but the quantum kernel cannot exploit that signal at any tested shot budget. Balanced-cohort metrics are benchmark values, not deployment-prevalence estimates.
This paper introduces a novel approach to quantifying quantum entanglement using entanglement entropy. Traditional entanglement measures, such as concurrence and entanglement of formation, often struggle to fully capture the intricate correlations present in entangled states. We propose a new definition of entanglement entropy based on the density matrix of the entangled system, explicitly accounting for the correlations between subsystems. This approach leverages information-theoretic entropy calculations to provide a more precise and comprehensive measure of entanglement strength and the degree of correlation. We demonstrate that this entropy-based metric offers advantages over existing measures, particularly in characterizing complex entangled states and distinguishing between different levels of entanglement. The core claim is that entanglement entropy can serve as a more accurate indicator of entanglement than existing metrics. The underlying mechanism centers around a density matrix-based entropy calculation, recognizing the interconnectedness within entangled states. The novelty lies in moving beyond purely state-based descriptions to incorporate a measure of subsystem correlations.
Abstract—Hybrid quantum–classical detectors are usuallyevaluated as monolithic alternatives to classical models, whichconflates two separable questions: whether the hard cases can beidentified, and whether a quantum model can resolve them. Weintroduce Q-RAD, a gated residual architecture that separatesthese questions by construction. A quantum-blind, quantumcost-free gating function routes only ambiguous transactions toa financial-transaction-specific quantum-kernel branch, whosescores are fused residually with a classical trunk under anexplicit finite-shot budget. We predeclare the architecture, aneight-comparison confirmatory family with Holm–Bonferronicorrection, a cost metric in quantum circuit executions, anda refutation posture, then evaluate on the Kaggle Credit-CardFraud dataset over ten repeated case-control splits under threeblocking validity gates, including exact-equality replay of previouslypublished results. The routing half of the architecturesucceeds: the gate attains 0.889 AUROC against realized classicalerrors and captures 77% of them within a 22% routing budget.The quantum half fails: handed exactly those hard cases, thequantum residual produced zero net corrections in nine of tenseeds and one net regression in the tenth. Q-RAD attains meancase-control test F1 0.916 against 0.918 for the classical trunkand 0.922 for an otherwise identical classical second opinion; thehypotheses of improvement over the classical trunk, over randomrouting, and over a classical second opinion are all refuted aftercorrection. We report this refutation as predeclared, and localiseit: ambiguity in this task is identifiable from classical signalsalone, but the quantum kernel cannot exploit that signal atany tested shot budget. Balanced-cohort metrics are benchmarkvalues, not deployment-prevalence estimates.Keywords—Quantum Machine Learning; Quantum KernelMethods; Selective Prediction; Learning to Defer; HybridQuantum-Classical Models; Financial Fraud Detection; ShotNoise; Negative Results