Skip to main content
QUICK REVIEW

[论文解读] MerLin: A Discovery Engine for Photonic and Hybrid Quantum Machine Learning

Cassandre Notton, Benjamin Stott|arXiv (Cornell University)|Feb 11, 2026
Neural Networks and Reservoir Computing被引用 0
一句话总结

MerLin 是一个开源框架,将线性光学电路的高保真仿真与 PyTorch、scikit-learn 集成,使光子量子模型的端到端可微训练成为可能,并在最先进的 QML 工作中进行基准测试。

ABSTRACT

Identifying where quantum models may offer practical benefits in near term quantum machine learning (QML) requires moving beyond isolated algorithmic proposals toward systematic and empirical exploration across models, datasets, and hardware constraints. We introduce MerLin, an open source framework designed as a discovery engine for photonic and hybrid quantum machine learning. MerLin integrates optimized strong simulation of linear optical circuits into standard PyTorch and scikit learn workflows, enabling end to end differentiable training of quantum layers. MerLin is designed around systematic benchmarking and reproducibility. As an initial contribution, we reproduce eighteen state of the art photonic and hybrid QML works spanning kernel methods, reservoir computing, convolutional and recurrent architectures, generative models, and modern training paradigms. These reproductions are released as reusable, modular experiments that can be directly extended and adapted, establishing a shared experimental baseline consistent with empirical benchmarking methodologies widely adopted in modern artificial intelligence. By embedding photonic quantum models within established machine learning ecosystems, MerLin allows practitioners to leverage existing tooling for ablation studies, cross modality comparisons, and hybrid classical quantum workflows. The framework already implements hardware aware features, allowing tests on available quantum hardware while enabling exploration beyond its current capabilities, positioning MerLin as a future proof co design tool linking algorithms, benchmarks, and hardware.

研究动机与目标

  • 推动对光子与混合QML在模型、数据集与硬件约束下的系统性、基准驱动探索。
  • 提供一个开放、可重复的框架,统一仿真、ML 工作流与光子 QML 的硬件访问。
  • 复现并验证 eighteen 种最先进的光子与混合 QML 工作,以建立共同基线。
  • 在 PyTorch 与 scikit-learn 生态系统中实现量子层的端到端可微训练。
  • 将 MerLin 定位为一个协同设计工具,连接算法、基准和硬件,以实现近端量子优势。

提出的方法

  • 将线性光学电路的强大仿真(SLOS)与 PyTorch 和 TorchScript 集成,以进行基于梯度的优化。
  • 在 PyTorch 中实现 QuantumLayer,暴露可训练的电路参数、批处理和与 ML 模型的互操作性。
  • 支持数据编码策略(角度编码和幅度编码)以及各种测量和探测器模型。
  • 提供面向硬件的设计,使用 MerlinProcessor 将部分计算卸载至光子量子处理单元及硬件约束。
  • 将光子量子模型嵌入现有 ML 流水线中,以实现消融实验、跨模态对比以及混合经典–量子工作流。
  • 采用以基准/可重复为先的实现与评估方法,在 MerLin 内实现并评估已发表的 QML 工作。
Figure 1 : MerLin architecture for photonic quantum machine learning. (A) Classical data encoding and photonic circuit configuration define the quantum model. (B) MerLin integrates PyTorch-based optimization with photonic-native execution through a logical-to-photonic bridge, differentiable quantum
Figure 1 : MerLin architecture for photonic quantum machine learning. (A) Classical data encoding and photonic circuit configuration define the quantum model. (B) MerLin integrates PyTorch-based optimization with photonic-native execution through a logical-to-photonic bridge, differentiable quantum

实验结果

研究问题

  • RQ1光子 QML 模型嵌入到标准 ML 工作流中的实际影响是什么?
  • RQ2一个统一框架是否能够在不同任务和体系结构上复现并基准大量光子和混合 QML 结果?
  • RQ3数据编码策略与硬件约束如何影响光子 QML 的性能与鲁棒性?
  • RQ4硬件感知的仿真在多大程度上能够引导算法与光子处理器的协同设计?
  • RQ5可重复、模块化的平台是否能够加速在 ML 领域实现近端量子优势的发现?

主要发现

  • MerLin 能复现十八种最先进的光子与混合 QML 工作,验证框架正确性并建立可复用的基线。
  • 硬件感知仿真与跨平台集成在若干场景下显著快于基于 Perceval 的实现。
  • Photon-native 模型在学习行为上常与基于门的模型相当,支持跨模态模型迁移与协同设计洞察。
  • 编码策略(角度编码 vs 幅度编码)在鲁棒性方面存在差异:幅度编码对对抗扰动更为敏感;角度编码相对更鲁棒。
  • 在 MerLin 中复现时,水库计算、QCNNs、QGANs 以及各种量子核和迁移学习方法显示出兼容的性能趋势,提升归因于实现细节和超参数。
(a) Parameterized linear photonic circuit with $m=3$ spatial modes and single-phase data encoding. The expectation value under PNR or threshold detection decomposes as $\sum_{\omega}c_{\omega}e^{i\omega x}$ , where $\omega$ depends on the photon number and $c_{\omega}$ on the trainable circuit and o
(a) Parameterized linear photonic circuit with $m=3$ spatial modes and single-phase data encoding. The expectation value under PNR or threshold detection decomposes as $\sum_{\omega}c_{\omega}e^{i\omega x}$ , where $\omega$ depends on the photon number and $c_{\omega}$ on the trainable circuit and o

更好的研究,从现在开始

从论文设计到论文写作,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。