Fig 1.
Quantum pipeline workflow overview.
Fig 2.
Quantum pipeline circuits example.
The first circuit (A) corresponds to the quantum k-NN, the second one (B) to the quantum binary classifier. In the case of the statevector modality, the final measurements are not present.
Table 1.
Quantum pipeline modalities (A), quantum binary classifier modalities (B), and baseline methods (C) considered.
Table 2.
Datasets properties (the dataset names are links that lead to the corresponding UCI pages).
Note: “qb.” stands for qubits.
Table 3.
Parameters of the experiments.
Fig 3.
Execution modalities comparison on 15 qubits datasets for the quantum pipeline.
Each point represents the accuracy obtained in a fold (or its average across runs).
Table 4.
Wilcoxon signed-rank test (α = 0.05) applied to the fold accuracy distributions shown in Fig 3.
The values reported in the table are the p-values obtained.
Table 5.
Average usage on dataset of the second model for the pipelines including the classical (or statevector) k-NN with cosine distance (A) and Euclidean distance (B).
The usage on dataset is 1 when the second model is always employed.
Fig 4.
Execution modalities comparison on 15 qubits datasets for the quantum binary classifier.
The 02_transfusion dataset is not present, and each point represents the accuracy obtained in a fold (or its average across runs). The p-value obtained by applying the Wilcoxon signed-rank test (α = 0.05) to the fold accuracy distributions is 0.016.
Fig 5.
Quantum pipeline—Quantum binary classifier comparison on common 15 qubits datasets.
Each point represents the accuracy obtained in a fold (or its average across runs); the k values refer only to the pipeline.
Table 6.
Wilcoxon signed-rank test (α = 0.05) applied to the fold accuracy distributions shown in Fig 5.
The values reported in the table are the p-values obtained.
Fig 6.
Dataset sizes (A) and distance metrics (B) comparisons.
In the dataset sizes comparison (A), each point represents the mean fold accuracy obtained on a dataset (or its average across runs); the pipeline comparisons include all k values. In the distance metrics comparison (B), the results obtained by the k-NN-based baseline methods (k-NN, k-NN + classifier, k-NN + SVM Gaussian, k-NN + SVM linear) on the 15 qubits datasets are taken into account; each point represents the accuracy obtained in a fold.
Table 7.
Wilcoxon signed-rank test (α = 0.05) applied to the mean fold accuracy distributions shown in Fig 6A (A). Same test applied to the fold accuracy distributions shown in Fig 6B (B).
Fig 7.
Quantum pipeline—Baseline methods comparison on 15 qubits datasets.
The pipeline modality is statevector—statevector, each point represents the accuracy obtained in a fold (or its average across runs), and the k-values refer only to the pipeline.
Fig 8.
Quantum pipeline—(k-NN-based) baseline methods comparison on 15 qubits datasets.
Each point in these plots represents the accuracy obtained in a fold.
Table 8.
Wilcoxon signed-rank test (α = 0.05) applied to the fold accuracy distributions shown in Fig 7.
The values reported in the table are the p-values obtained.
Table 9.
Wilcoxon signed-rank test (α = 0.05) applied to the fold accuracy distributions shown in Fig 8.
The values reported in the table are the p-values obtained.