Towards Defect Prediction for Quantum Software*

Software defects are faults or bugs within a program that can lead to incorrect or unexpected outcomes. Efficiently allocating software quality assurance (SQA) resources to components with a higher likelihood of defects, based on software defect prediction (SDP) models, can save significant effort. Although SDP has been extensively studied in classical software, its applicability to quantum software remains unexplored. Defect prediction for quantum software presents unique challenges, including the susceptibility to quantum-specific defects arising from quantum coding conventions and the limited size of available datasets. To address these issues, we propose QDP-FSL, an SDP model using a pre-trained code model to capture the semantics of quantum software code, and applying few-shot learning (FSL) to learn effectively from a small number of defective samples. Results show that QDP-FSL outperforms baseline methods that rely on static analysis. This work lays the foundation for future research in defect prediction for quantum software engineering and outlines potential directions for further improvement.