Quantum-Integrated AI
量子集成AI
This program studies hybrid classical-quantum workflows in which small quantum components serve as constrained decision elements within larger classical systems. The focus is on bounded sampling integration, where quantum resources contribute to specific decision points under defined error margins rather than replacing classical computation entirely.
本项目研究混合经典-量子工作流，其中小型量子组件作为大型经典系统中的约束决策元素。重点在于有界采样集成，量子资源在定义的误差范围内为特定决策点做出贡献，而非完全替代经典计算。
All quantum integration work operates under the assumption that current and near-term quantum hardware is noisy and limited. Research proceeds by defining constraint surfaces that make this noise budget explicit and traceable.
所有量子集成工作均基于当前和近期量子硬件存在噪声且有限的假设。研究通过定义使噪声预算明确可追溯的约束面来推进。
Scope
研究范围
- Bounded sampling integration: embedding quantum sampling operations into classical inference pipelines at defined decision points, with explicit error bounds and fallback paths. 有界采样集成：在定义的决策点将量子采样操作嵌入经典推断管道，具有明确的误差边界和回退路径。
- Hybrid workflow orchestration: coordination protocols for systems that mix classical and quantum resources, including queue management, resource allocation, and result reconciliation. 混合工作流编排：混合经典和量子资源系统的协调协议，包括队列管理、资源分配和结果协调。
- Quantum constraint surfaces: formal definitions of the operating bounds for quantum components, including qubit counts, gate fidelities, and coherence time requirements. 量子约束面：量子组件运行边界的形式化定义，包括量子比特数、门保真度和相干时间要求。
- Decoherence-aware scheduling: task scheduling algorithms that account for coherence time limitations and schedule quantum operations to maximize useful computation within available windows. 退相干感知调度：考虑相干时间限制的任务调度算法，在可用窗口内安排量子操作以最大化有效计算。
- Error-bounded inference paths: inference architectures where quantum-derived results carry explicit confidence intervals and propagate uncertainty through downstream classical processing. 误差有界推断路径：量子推导结果携带明确置信区间并将不确定性传播到下游经典处理的推断架构。
Evaluation Harness
评估框架
- Error margin verification: quantum sampling results are compared against classical baselines to verify that observed error rates remain within the declared constraint surface. 误差范围验证：量子采样结果与经典基准进行比较，以验证观察到的误差率保持在声明的约束面内。
- Decoherence budget tracking: monitoring of coherence time utilization across all quantum operations, with alerts when operations approach or exceed their allocated coherence windows. 退相干预算跟踪：监控所有量子操作的相干时间利用率，当操作接近或超过分配的相干窗口时发出警报。
- Hybrid throughput measurement: end-to-end latency and throughput metrics for workflows that include quantum components, measured against equivalent all-classical implementations. 混合吞吐量测量：包含量子组件的工作流的端到端延迟和吞吐量指标，与等效的全经典实现进行对比测量。
Open Questions
开放问题
- At what problem sizes does bounded quantum sampling provide measurable advantage over classical sampling for the specific decision tasks studied in this program? 对于本项目研究的特定决策任务，有界量子采样在何种问题规模下提供可测量的优于经典采样的优势？
- Can constraint surfaces for quantum components be composed with constraint surfaces for classical components in a way that preserves formal guarantees? 量子组件的约束面能否与经典组件的约束面以保持形式化保证的方式进行组合？
- What is the minimum gate fidelity threshold below which decoherence-aware scheduling cannot recover useful computation within the defined error bounds? 退相干感知调度无法在定义的误差边界内恢复有效计算的最低门保真度阈值是多少？
Lab Notes
实验笔记
Initial bounded sampling experiments using 7-qubit circuits show that for the specific combinatorial decision tasks in yl-qi-003, quantum sampling matches classical performance at approximately 10^4 candidate evaluations and provides measurable improvement at 10^5. These results are preliminary and contingent on the gate fidelity levels available in the test hardware (approximately 99.2% single-qubit, 97.8% two-qubit).
使用7量子比特电路的初始有界采样实验表明，对于 yl-qi-003 中的特定组合决策任务，量子采样在约10^4候选评估时匹配经典性能，并在10^5时提供可测量的改进。这些结果是初步的，取决于测试硬件中可用的门保真度水平（约99.2%单量子比特，97.8%双量子比特）。
The hybrid workflow orchestrator now supports dynamic fallback: when quantum hardware queue depth exceeds a defined threshold, affected tasks are automatically rerouted to classical sampling with adjusted confidence intervals. This prevents queue-induced latency from propagating through the pipeline. Fallback rates in the current deployment average 8-12% of scheduled quantum operations.
混合工作流编排器现支持动态回退：当量子硬件队列深度超过定义阈值时，受影响的任务自动重新路由到经典采样并调整置信区间。这防止了队列引起的延迟在管道中传播。当前部署中的回退率平均为计划量子操作的8-12%。
量子边界与约束面：本组的工作假设是，量子组件的实用价值不在于计算加速，而在于提供经典方法难以有效采样的约束面。这一观点将量子资源从通用计算工具重新定位为特定决策拓扑的采样装置。当前的形式化工作集中在定义哪些约束面拓扑结构适合量子采样，以及如何量化这种适合性。初步分类将约束面分为三类：量子有利、量子中性和量子不利。
Citations
参考文献
-
yl-qi-ref-001
"Bounded Sampling in Hybrid Classical-Quantum Workflows." Internal Working Paper, Quantum Workflows Unit. 2025. -
yl-qi-ref-002
"Decoherence-Aware Task Scheduling: Formal Bounds and Practical Heuristics." Technical Report yl-tr-031. Yueqian Labs. 2025. -
yl-qi-ref-003
"Constraint Surface Composition for Mixed Classical-Quantum Systems." Preprint, submitted to Workshop on Formal Methods in Quantum Computing. 2025. -
yl-qi-ref-004
"Error Propagation in Hybrid Inference Paths: Empirical Measurements on 7-Qubit Circuits." Lab Report yl-qi-lr-007. 2024.