qml-mutation-testing

name	qml-mutation-testing
description	Systematic mutation testing methodology for quantum machine learning models. Use when testing, validating, or assessing robustness of quantum ML models, variational quantum circuits, and QNNs.
category	quantum-ml

QML Mutation Testing

Description

Systematic mutation testing methodology for quantum machine learning (QML) models. Provides structured approaches for testing correctness and robustness of quantum circuits through controlled mutations.

When to Use

• Testing quantum machine learning model correctness
• Validating robustness of variational quantum circuits
• Assessing QNN resilience to parameter perturbations
• Building test suites for quantum algorithms

Mutation Operators for QML

Gate-Level Mutations

• Gate replacement: Replace a gate with a different gate type (e.g., RX to RY)
• Parameter perturbation: Modify rotation angles by small amounts
• Gate deletion: Remove gates from the circuit
• Gate insertion: Add extra gates at random positions
• Gate order swap: Change the ordering of consecutive gates

Circuit-Level Mutations

• Entanglement modification: Change CNOT/CZ gate targets or controls
• Qubit permutation: Swap qubit assignments in the circuit
• Depth variation: Increase or decrease circuit depth
• Measurement mutation: Change measurement basis (X, Y, Z)

Parameter-Level Mutations

• Initial parameter perturbation: Modify starting parameters
• Learning rate mutation: Change optimization step sizes
• Regularization mutation: Add or remove regularization terms

Testing Workflow

1. Define mutant set: Select mutation operators relevant to the QML model
2. Generate mutants: Apply each mutation to create variant circuits
3. Evaluate mutants: Run each mutant through the same training/evaluation pipeline
4. Compute kill rate: Determine what fraction of mutants are detected (degraded performance)
5. Analyze survivors: Study which mutations the model is insensitive to

Metrics

• Mutation score: Fraction of mutants killed (higher = better test suite)
• Sensitivity profile: Which mutation types most affect model performance
• Robustness index: Correlation between mutation severity and performance degradation
• Equivalent mutants: Mutants that behave identically to original

Pitfalls

• Quantum noise on real hardware can mask mutation effects - use simulators for clean testing
• Some mutations may produce functionally equivalent circuits
• Mutation size matters: too small and no effect, too large and trivially detected
• Account for quantum measurement stochasticity when comparing mutant outputs

Related QML Validation Patterns

Mutation testing is one part of a broader QML quality assurance workflow. Complementary patterns:

• Robustness analysis: Evaluate QNNs under noise and adversarial perturbations (see ml-quantum-error-correction skill, references/qml-model-validation.md)
• Certified training: Quantum Interval Bound Propagation (IBP) for adversarial robustness guarantees
• Hardware readiness: Circuit compilation, error mitigation, deployment checklist for real quantum backends

References

• Andrews, Mishra. "Efficient Mutation Testing of Quantum Machine Learning Models" (arXiv:2605.00107, 2026)