INTERPRETING QUANTUM CIRCUIT LEARNING WITH QPERT: A STEP TOWARD TRUSTWORTHY QUANTUM AI

Smriti Bajaj, Ramesh Doddaiah, Garima Jain

Keywords: Explainable Quantum AI, XAI, Quantum computing, Quantum circuit learning

Abstract: Quantum Circuit Learning (QCL) presents a promising hybrid computational framework that combines the representational capacity of parameterized quantum circuits (PQCs) with classical optimization techniques for solving machine learning problems. However, the opaque nature of QCL models limits their adoption in domains requiring transparency and accountability. In this work, we introduce quantum perturbation (QPERT), a novel perturbation-based explainability approach tailored for QCL. QPERT generates a saliency mask by quantifying the importance of input features for a given instance while preserving key quantum properties such as entanglement and superposition. We evaluate QPERT in explaining a hybrid quantum-classical architecture trained on the Iris dataset. Comparative analysis against established explainability techniques, including SHAP and LIME, highlights QPERT's effectiveness in delivering interpretable insights into quantum model behavior. Our results demonstrate the feasibility of interpretable quantum learning and offer practical guidance for integrating explainability into quantum-classical pipelines.

Primary Area: interpretability and explainable AI

Submission Number: 6953