面向金融收益的可解释量子生成器:基于问题感知的量子电路设计
Toward Interpretable Quantum Generators for Financial Returns through Problem-Informed Circuit Design
AI总结:
该研究针对金融收益生成的需求,提出问题感知的量子电路设计框架,在混合Wasserstein量子生成对抗网络中验证了三种量子生成器架构,实现了紧凑可解释的量子生成器,为量子电路设计提供了新方向。
摘要:
量子电路作为函数近似器时,常缺乏可解释性,且在表达能力与可训练性之间存在根本权衡。我们提出一种面向金融收益生成的问题感知框架,利用关系结构和对称性启发的参数共享,指导构建具有少量可训练参数、算子角色可经验验证的量子电路。我们在混合Wasserstein量子生成对抗网络(WGAN)中评估三种量子生成器架构,即ZZ、定向ZZ-CRX和结构化时间演化,用于生成16步比特币收益窗口。在1至4层、1至8个可训练系数的电路中,这些生成器能恢复有意义的边际分布和时间分布特征,且单层配置与更深层配置相比仍具竞争力。在匹配深度下,独立训练的ZZ-CRX配置的中位数杠杆率相对ZZ呈显著下降,这与受控旋转结构的预期定向作用一致。少量系数还支持直接的损失景观分析,揭示了边际目标与时间目标之间相互竞争的参数偏好,有助于解释为何难以找到同时在两个目标上表现良好的参数设置。这些结果初步证明,结构假设可指导紧凑量子生成器的设计,并将电路架构与可观测的统计行为关联起来,为开发更系统、参数效率更高的量子电路构建方法提供了动力。
英文摘要:
When used as function approximators, quantum circuits often lack interpretability and face a fundamental trade-off between expressivity and trainability. We propose a problem-informed framework for financial return generation that uses relational structure and symmetry-inspired parameter sharing to guide the construction of quantum circuits with few trainable parameters and empirically testable operator roles. We evaluate three quantum generator architectures, referred to as ZZ, directed ZZ-CRX, and structured time evolution, within a hybrid Wasserstein quantum generative adversarial network trained to generate 16-step Bitcoin return windows. Across circuits with one to four layers and one to eight trainable coefficients, these generators recover meaningful marginal and temporal distributional features, while single-layer configurations remain competitive with deeper variants. Across matched depths, independently trained ZZ-CRX configurations exhibit a predominantly downward shift in median leverage relative to ZZ, consistent with the intended directional role of the controlled-rotation construction. The small number of coefficients also enables direct landscape analysis, revealing competing parameter preferences between marginal and temporal objectives and helping to explain the difficulty of finding parameter settings that perform well on both. These results provide an initial demonstration that structural assumptions can guide the design of compact quantum generators and link circuit architecture to observable statistical behaviour, motivating more systematic and parameter-efficient approaches to quantum circuit construction.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。