Abstract Early-stage venture screening requires navigating non-linear compensatory trade-offs that linear scoring models struggle to capture. We propose HQDIS, a parsimonious 4-qubit Variational Quantum Circuit (13 parameters) for a stylized version of this problem, conceptually motivated by a published study of Canadian venture investors (Bachher, Díaz de León, and Guild, 1999). Calibration uses a synthetic N = 45 benchmark (Benchmark A) authored by the researchers, not real historical decisions. We introduce a monotonic angular-scaling protocol (π/4) that provably prevents a phase-inversion aliasing failure in a common feature map. Under Repeated Stratified 5-Fold Cross-Validation (100 repeated resampling runs over the same 45 cases, not 100 independent samples), HQDIS attains 80.33% (± 2.72%, 95% CI) accuracy, statistically indistinguishable from Random Forest and Gradient Boosting (Nadeau–Bengio p > 0.7); Logistic Regression attains a nominally higher 91.56% (p = 0.133, d=-0.77), a gap our tests lack power to confirm at N = 45. Ablation shows a nominal 7-point gap and a medium effect size (d ≈ 0.51) between the full circuit and variants lacking ansatz entanglement, but the formal paired test does not reach significance (p ≈ 0.32) at this sample size. We execute Benchmark A on IBM's 156-qubit Heron processor (ibm_fez), obtaining 93.33% hardware accuracy and 97.78% concordance with simulation (κ = 0.9552; r = 0.9943) — a measure of hardware-simulation fidelity, not predictive validity. A counterfactual sweep on one boundary case demonstrates a reproducible auditability mechanism. In a second, explicitly synthetic benchmark (Benchmark B) built to favor non-linear rules, Logistic Regression significantly outperformed HQDIS in accuracy (paired p = 0.025) across 10 seeds, while HQDIS showed a non-significant nominal AUC-ROC advantage (paired p = 0.141); Random Forest significantly exceeded HQDIS on both metrics. We report this mixed result in full. We discuss the resulting, circumscribed implications for quantum decision-support architectures and report Benchmark A's synthetic construction as a central scope limitation.
Abstract Early-stage venture screening requires navigating non-linear compensatory trade-offs that linear scoring models struggle to capture. We propose HQDIS, a parsimonious 4-qubit Variational Quantum Circuit (13 parameters) for a stylized version of this problem, conceptually motivated by a published study of Canadian venture investors (Bachher, Díaz de León, and Guild, 1999). Calibration uses a synthetic N = 45 benchmark (Benchmark A) authored by the researchers, not real historical decisions. We introduce a monotonic angular-scaling protocol (π/4) that provably prevents a phase-inversion aliasing failure in a common feature map. Under Repeated Stratified 5-Fold Cross-Validation (100 repeated resampling runs over the same 45 cases, not 100 independent samples), HQDIS attains 80.33% (± 2.72%, 95% CI) accuracy, statistically indistinguishable from Random Forest and Gradient Boosting (Nadeau–Bengio p > 0.7); Logistic Regression attains a nominally higher 91.56% (p = 0.133, d=-0.77), a gap our tests lack power to confirm at N = 45. Ablation shows a nominal 7-point gap and a medium effect size (d ≈ 0.51) between the full circuit and variants lacking ansatz entanglement, but the formal paired test does not reach significance (p ≈ 0.32) at this sample size. We execute Benchmark A on IBM's 156-qubit Heron processor (ibm_fez), obtaining 93.33% hardware accuracy and 97.78% concordance with simulation (κ = 0.9552; r = 0.9943) — a measure of hardware-simulation fidelity, not predictive validity. A counterfactual sweep on one boundary case demonstrates a reproducible auditability mechanism. In a second, explicitly synthetic benchmark (Benchmark B) built to favor non-linear rules, Logistic Regression significantly outperformed HQDIS in accuracy (paired p = 0.025) across 10 seeds, while HQDIS showed a non-significant nominal AUC-ROC advantage (paired p = 0.141); Random Forest significantly exceeded HQDIS on both metrics. We report this mixed result in full. We discuss the resulting, circumscribed implications for quantum decision-support architectures and report Benchmark A's synthetic construction as a central scope limitation.
This study investigates classification performance and trainability of variational quantum classifiers on near-term noisy quantum hardware across circuit depth and qubit count regimes within the context of quantum computing and machine learning, an area of growing scientific importance given its implications for hybrid quantum-classical algorithm design and practical NISQ hardware benchmarking for near-term quantum advantage assessment. Using gradient-based parameter optimization of parameterized quantum circuits with hardware noise modeling via density matrix simulation and physical execution on IBM Eagle processors, we examine expressibility of parameterized circuits enabling nonlinear decision boundaries in feature Hilbert space, with trainability constrained by barren plateau gradients at large qubit counts in 96 circuit architectures (4 qubit counts x 4 depths x 6 ansatz families) evaluated across 5 classification datasets with 20 random initialization seeds each drawn from IBM Quantum Eagle (127-qubit) and Falcon (27-qubit) processors with noise mitigation via zero-noise extrapolation. Results indicate that 10-qubit, 4-layer VQC achieves 84.2% classification accuracy on the best benchmark, within 3.8 percentage points of classical SVM, but gradient variance decays exponentially with qubit count (variance halving per 2 added qubits), confirming barren plateau scaling (p < 0.001), with 84.2% accuracy, 3.8 pp gap to classical SVM baseline as the primary quantitative benchmark. Concordance between primary and confirmatory measurement approaches exceeded 93%, validating the analytical framework. These findings contribute empirically to quantum computing and machine learning and carry actionable implications for the design of programs and policies targeting hybrid quantum-classical algorithm design and practical NISQ hardware benchmarking for near-term quantum advantage assessment.
Abstract We introduce Quantum Diagnostic Intelligence (QDI), a five-layer decision-support architecture for inferring the operational state of quantum processing units from circuit-level measurements. Pulse-level calibration metrics — coherence times T₁ and T₂ measured through single-qubit Rabi and Ramsey sequences — are the dominant industry standard for certifying quantum hardware health. We demonstrate, through 191 circuit executions on three 156-qubit IBM Heron r2 processors, that circuit-level measurements contain diagnostic information that T₁/T₂ may not capture. On ibm_marrakesh between July 3 and July 5, 2026, IBM-reported T₂ improved from 303 to 340 µs (+ 12%), while circuit-level GHZ scaling yielded α = 3.811, a QDI health score of 37.8/100 [95% CI: 7.6, 67.9], and a circuit infidelity increase of 920% at n = 2 qubits. The QDI architecture formalizes this observation through four operators: a diagnostic state classifier D(χ), a health score H(χ), a recommendation generator R(χ), and a drift forecast F(χ, t). Rules trained on ibm_kingston correctly classified ibm_fez and both ibm_marrakesh sessions without retuning (3/3 held-out tests; n = 4 total — feasibility demonstration only). We identify three scientific limitations of the current framework: operators rely on expert-designed rules rather than learned statistical models; ground truth (algorithm success/failure) has not yet been collected; and diagnostic state labels have not been validated by clustering analysis. Prospective validation on n ≥ 50 independent sessions is required before statistical claims about generalization can be made. The dataset, code, and QDI system are openly available.
Hardware validation on IBM’s 133-qubit Heron processor demonstrates that a quantum neural architecture can reliably distinguish neural configurations with 99.3% discrimination quality. The A-Gate circuit encodes excitatory-inhibitory dynamics from EEG signals using paired qubits, achieving O(M) gate complexity for M-channel correlation encoding versus O(M2) classical computation on the symmetric positive definite (SPD) manifold. Transpilation to native CZ gates follows the predicted linear law over the implemented range ( 14.1M−17.5 , R2>0.99 for M≤8 ); extending to M=32 reveals super-linear growth from topology-induced routing overhead, which we report as a hardware limitation distinct from the architecture’s O(M) logical complexity. Evaluation on the CHB-MIT Scalp EEG Database using leave-one-subject-out (LOSO) cross-validation across 22 subjects reveals a critical encoding bottleneck: the quantum fidelity classifier achieves 53.4% AUC compared to 62.5% for an 8-channel classical baseline. Notably, per-subject AUC ranges from 0.183 to 1.000, with sub-chance performance in some cases suggesting what we term manifold polarity, a hypothesized patient-specific geometric orientation that may cause cross-subject template mismatch. The gap between quantum and classical performance indicates that mapping EEG features to circuit parameters, not the A-Gate architecture itself, limits classification. We provide complexity analysis with corrections to common quantum-advantage overclaims, and we are explicit that the present results establish hardware feasibility and the location of the performance bottleneck rather than clinical utility.
Abstract Claims of quantum advantage in machine learning often rest on classical comparators that lack the same features, on simulation alone, and on unreported hardware cost. QBioBench is a benchmark of quantum machine learning on peripheral biosignals that holds quantum and classical models to the same features and reports calibration and hardware cost. It comprises three clinical tasks, 12-lead ECG diagnosis (PTB-XL), single-lead apnea detection (Apnea-ECG), and wearable stress detection, each reduced to six to eight shared dimensions. A factorial grid of projected and fidelity quantum kernels and variational classifiers is screened in noisy simulation, and the promoted configurations are evaluated in an exploratory study on an IBM Nighthawk r2 processor on the full test pools of 510 apnea minutes and 512 PTB-XL records. Every prediction is recovered from the device records, intervals resample recordings or patients, and classical baselines are trained on the same units and tested on the same executed inputs. No quantum model reaches its matched support vector machine in ROC-AUC on either primary task, and none is separably better in balanced accuracy. The angle-encoded fidelity and projected kernels reach ROC-AUC 0.697 and 0.693 on apnea, paired differences of − 0.127 and − 0.119 to the support vector machine, and both are classical by construction, since the entangling layer of the encoding cancels from the fidelity kernel and leaves closed-form local expectations in the projected kernel. A variational classifier comes closest, reaching 0.797 on apnea (difference − 0.012 , 95% interval [ − 0.052 , + 0.027 ] ), and falls to chance without its two-qubit gates. Test sets of the size that circuit budgets favor are unreliable, as random 80-minute subsets of the apnea pool give the fidelity kernel a central 95% range of ROC-AUC from 0.589 to 0.797. Readout mitigation changed ROC-AUC by at most 0.015.
Despite rapid progress in quantum optimization, broad real-hardware benchmarks comparing multiple algorithmic families across diverse combinatorial problems under a common protocol remain limited. We benchmark gate-based quantum optimization on four NP-hard problems: multi-dimensional knapsack (MDKP), maximum independent set (MIS), quadratic assignment (QAP), and market-share (MSP). We study VQE, CVaR-VQE, standard, multi-angle, and warm-start QAOA, together with qubit-efficient PCE and QRAO, on IBM Heron r1/r2 processors using resilience-level-2 mitigation. To our knowledge, this includes the first real-hardware QRAO results and the first multi-problem PCE hardware benchmark.
Across 247 method-instance combinations, we report transpiled circuit size, hardware outcomes, and an independent-error gate-count fidelity proxy, Fest. For MDKP and MIS, an empirical operating point near Fest≈0.1, corresponding to about 770 two-qubit gates at the median Heron-r2 CZ error rate, marks the onset of noise-dominated execution. QAP exposes a separate bottleneck: dense one-hot encodings and an exponentially sparse feasible manifold, with feasible fraction 10!/2100 at n=10; no tested hardware method produces a feasible assignment. Compiled QAOA-family circuits are generally noise dominated, and a matched uniform-random control shows that most feasible low-fidelity outcomes fall within the random range, apart from one finite-sample MIS warm-start exception. A SWAP-aware, fractional-gate, Nighthawk-topology compilation counterfactual reduces two-qubit counts but leaves all circuits below Fest=10−3. These conclusions apply to the tested implementations rather than QAOA in general. Qubit-efficient methods extend runnable instance sizes, but only within the empirical fidelity budget.
Hybrid quantum-classical neural networks offer a parameter-efficient path for clinical prediction, yet the field lacks reproducible methodologies for architectural design. Most current models rely on ad hoc circuit choices, complicating replication and comparison. This study introduces a generalizable Hybrid Quantum-Classical Neural Network (HQCNN) framework that replaces trial-and-error design with a principled Bayesian-surprise-guided methodology. Evaluated on the Wisconsin Diagnostic Breast Cancer dataset (n = 569), the framework employs a four-component PCA pipeline feeding a 4-qubit parameterized quantum circuit with two variational layers, integrated within a classical neural pipeline. The model was benchmarked against tuned Support Vector Machine, Random Forest, XGBoost, and Multi-Layer Perceptron baselines under identical 5-fold stratified cross-validation with nested GridSearchCV. The HQCNN achieved 96.49% ± 1.24% accuracy and 99.51% ± 0.38% AUC, outperforming a structurally comparable MLP while using 11.27% fewer trainable parameters (441 versus 497). A circuit-depth ablation identified two variational layers as optimal, consistent with barren-plateau dynamics. KL divergence scores of 0.925, 0.804, and 0.653 nats quantified the epistemic informativeness of competitive accuracy, optimal shallow depth, and parameter efficiency, respectively, while the AI2 AutoDiscovery platform independently validated preprocessing choices post hoc. These results indicate that the primary near-term value of hybrid models in healthcare lies in empirical parameter efficiency rather than raw accuracy gains. Fewer parameters reduce overfitting risk on small medical datasets, lower deployment costs, and produce models that are easier to audit for clinical governance. The Bayesian-surprise methodology finally provides the reproducible, principled design framework that the field has long lacked.