Interpretable Quantum Advantage in Neural Sequence Learning

Quantum neural networks have been widely studied in recent years, given their potential practical utility and results regarding ability to efficiently express certain classical data. However, analytic date rely on assumptions arguments from complexity theory. Due this, there is little intuition as the source of expressive power quantum or for which classes data any advantage can be reasonably expected hold. Here, we study relative between a broad class network sequence models recurrent based Gaussian operations with non-Gaussian measurements. We explicitly show that contextuality an unconditional memory separation expressivity two model classes. Additionally, are able pinpoint this separation, use performance our introduced standard translation set exhibiting linguistic contextuality. In doing so, demonstrate outperform state art even practice.