This study introduces a framework for quantitatively comparing the generalization performance of quantum and classical generative models to determine practical quantum advantage (PQA). The framework uses a sample-based approach, making it model-agnostic. Quantum Circuit Born Machines (QCBMs) are compared against Transformers, Recurrent Neural Networks, Variational Autoencoders, and Wasserstein Generative Adversarial Networks. Results suggest QCBMs are more efficient in data-limited regimes, a highly desirable feature for real-world applications with scarce data.

Mohamed Hibat-Allah, Marta Mauri, Juan Carrasquilla, Alejandro Perdomo-Ortiz