A new arXiv preprint describes an approach to scaling quantum natural language processing for Hindi by automating pregroup supertagging. The work builds on prior Hindi-specific pregroup grammars that map grammatical structure into diagrammatic and quantum circuit representations. Automating the tagging step is intended to reduce the manual effort needed to apply these grammars at larger scale.
Reinforcement learning can automate the design of quantum circuits, but performance degrades as the search space grows and rewarding circuit trajectories become rare. This arXiv paper proposes a generative replay technique that synthesizes new training samples instead of only reusing previously observed transitions, addressing the resulting data shortage. The authors report that this approach improves the scalability of quantum architecture search.
A new arXiv paper tackles how to schedule quantum circuits across systems that combine high-performance computing with several quantum processing units, where the QPUs differ in size, connectivity, native gate sets, and noise. The proposed approach factors in these hardware differences when assigning circuits, aiming to maximize fidelity on today's noisy devices where errors compound during compilation.