superpositions
    Log inStart free trial

    QSVM
    Quantum Support Vector Machine

    Quantum kernel estimation and classification on NISQ backends: transparent, reproducible, and benchmarked on Superpositions Studio.

    Reproducible quantum kernels
    Downloadable code & benchmarks
    Classical SVM baselines included
    Transparent metrics & results
    Run QSVM Simulation

    Overview

    A QSVM uses quantum feature maps (kernels) to embed classical data into a high-dimensional Hilbert space, producing kernel values via quantum circuits. A classical SVM is then trained on the quantum-computed kernel matrix. Potential advantages arise when the induced quantum kernel is hard to simulate classically, yielding better data separability for some datasets.

    Why It Matters

    Kernel methods remain powerful for small-to-medium datasets and tricky decision boundaries. Quantum kernels can, in principle, offer richer feature spaces, potentially improving classification margins on certain problems.

    How QSVM Works

    A simple four-step process to leverage quantum kernels for enhanced classification

    01

    Choose a quantum feature map

    Select Uφ(x) that encodes data point x into a quantum state.

    02

    Estimate kernel entries

    Calculate Kij = ⟨φ₀|Uφ(xⱼ)|xᵢ⟩ using a quantum backend.

    03

    Train classical SVM

    Use the quantum kernel matrix to train your SVM classifier.

    04

    Validate and tune

    Test generalization; tune feature map depth and regularization.

    Real-World Applications

    Where quantum kernels provide tangible advantages

    Security

    Anomaly Detection

    Classification for structured signals and outlier identification

    Efficiency

    Small-Data Regimes

    Optimal performance where kernel methods excel

    Advanced

    Complex Separations

    Datasets where quantum feature maps induce useful separations

    Innovation

    Research Prototypes

    Quantum-enhanced ML research and experimentation

    Strengths & Limitations

    Strengths

    • Strong baseline approach for NISQ-era quantum ML
    • Interpretable via kernels and margins
    • Works well with small datasets
    • Proven classical foundation with quantum enhancement

    Limitations

    • Kernel matrix estimation is expensive (quadratic in dataset size)
    • Noise impacts kernel fidelity
    • No guaranteed advantage over tuned classical kernels
    • Problem-dependent performance

    Benchmarking & Verification

    Compare quantum kernels to RBF/Polynomial baselines; report learning curves, margins, ROC, and calibration plots. Seed-controlled runs and versioned artifacts ensure reproducibility.

    Hardware & Requirements

    QubitsDepends on feature map width
    Circuit DepthGrows with map complexity; shallow maps more noise-tolerant
    Shots1k–100k depending on kernel variance
    BackendSimulator / small NISQ devices
    ScalabilityO(n²) kernel entries: grows with dataset size

    Note: Kernel estimation cost grows with dataset size (O(n²) kernel entries)

    Proof-of-Concept

    Real experimental results demonstrating QSVM performance

    Task

    Binary classification on a synthetic 2D dataset

    Feature Map

    Data-reuploading with 1–2 layers

    MetricsAUC, F1 Score, Margin, Runtime, Shot Budget
    OutcomeComparable to strong classical kernels on small datasets

    FAQ

    Common questions about QSVM implementation and performance

    Do QSVMs always outperform classical SVMs?

    No. Benefits are problem-dependent and sensitive to feature map design and noise.

    Can QSVM run on real hardware?

    Yes, for small datasets/feature maps; simulations scale more easily.

    How do I choose a feature map?

    Start shallow and problem-inspired; increase depth cautiously to avoid noise and barren plateaus.

    Related pages

    Algorithms, applied use cases and benchmarks connected to this page

    Healthcare

    Breast Cancer Classification with QSVM

    Reproducible QSVM breast-cancer classification on FNA data, including an LS-QSVM with HHL linear-system variant benchmarked against classical SVM.

    Manufacturing

    Quantum Quality Control in Additive Manufacturing with QSVM

    Detect additive manufacturing anomalies and classify melt-pool defect shapes with QSVM, reproducible benchmarks, and a classical baseline comparison.

    Manufacturing

    Steel Plate Fault Detection with QSVM

    Detect steel plate faults with QSVM, reproducible industrial benchmarks, downloadable reports, and inspection-ready classification workflows.

    Comparisons

    QSVM vs Classical SVM for Classification

    Compare QSVM with classical SVM on a 569-sample breast-cancer benchmark with 30 FNA features, 4 qubits, 2 layers, 96% accuracy, and 0.99 ROC-AUC.

    Comparisons

    HQNN vs Classical Neural Network

    Compare hybrid quantum neural networks with classical MLP and CNN baselines on parameters, training time, and accuracy across three benchmarked use cases.

    Ready to Run QSVM?

    Build quantum kernels, compare against classical baselines, and export reproducible code and comprehensive reports.

    Start QSVM Simulation

    Powered by Superpositions Studio: transparent, reproducible quantum computing