This paper introduces the Quantum Generative Diffusion Model (QGDM), a fully
quantum-mechanical model for generating quantum state ensembles, inspired by
Denoising Diffusion Probabilistic Models. QGDM features a diffusion process
that introduces timestep-dependent noise into quantum states, paired with a
denoising mechanism trained to reverse this contamination. This model
efficiently evolves a completely mixed state into a target quantum state
post-training. Our comparative analysis with Quantum Generative Adversarial
Networks demonstrates QGDM's superiority, with fidelity metrics exceeding 0.99
in numerical simulations involving up to 4 qubits. Additionally, we present a
Resource-Efficient version of QGDM (RE-QGDM), which minimizes the need for
auxiliary qubits while maintaining impressive generative capabilities for tasks
involving up to 8 qubits. These results showcase the proposed models' potential
for tackling challenging quantum generation problems.

此研究介绍了量子生成扩散模型（QGDM），一种完全的量子力学模型，用于生成量子态系列，灵感来自去噪扩散概率模型。QGDM 通过引入时间步骤相关的噪声进入量子态，并配对训练以逆转这种污染的降噪机制，高效地将完全混合态演化成目标量子态。与量子生成对抗网络的比较分析证明了 QGDM 的优越性，在涉及 4 个量子比特的数值模拟中，拟真度指标超过 0.99。此外，我们提出了一种资源高效的 QGDM 版本（RE-QGDM），其在涉及 8 个量子比特的任务中减少了对辅助量子比特的需求，同时保持了令人印象深刻的生成能力。这些结果展示了所提出模型在应对具有挑战性的量子生成问题方面的潜力。