Many problems in science, engineering, and finance involve generating samples from complex probability distributions. Classically, such samples are usually the end of the story: one draws random outcomes and then analyzes them. Quantum samples are more powerful. By encoding many possible outcomes coherently in a quantum state, they can be passed directly into other quantum algorithms for tasks such as estimating probabilities, detecting rare events, optimization, learning, and simulating complex systems. This makes efficient quantum sampling a key building block for future quantum technologies.
Here, we show how to make such quantum samples less memory-intensive to generate. The work introduces a form of lossy quantum dimension reduction for stochastic processes: instead of storing every detail of the past perfectly, the quantum model keeps only the information most relevant for accurately reproducing future behaviour. This can reduce the required quantum memory below both classical models and previous quantum constructions, while still producing high-quality samples. By lowering the resource cost of preparing quantum samples from processes with long-range temporal correlations, the method helps make these samples more practical as inputs for downstream quantum algorithms.
- Dimension Reduction in Quantum Sampling of Stochastic Processes
Chengran Yang, Marta Florido-Llinàs, Mile Gu, and Thomas J. Elliott
npj Quantum Information 11, 34
