量子机器学习概念
使用集合让一切井井有条
根据您的偏好保存内容并对其进行分类。
Google 的量子超越传统实验使用了 53 个嘈杂量子位，证明一台量子计算机只需 200 秒就可以完成采用现有算法的最大传统计算机需要大约 10,000 年才能完成的一项计算。这标志着嘈杂中型量子 (NISQ) 计算时代正式开启。在未来几年中，具有数十乃至数百个嘈杂量子位的量子设备有望成为现实。
量子计算
量子计算依靠量子力学的属性来计算传统计算机无法解决的问题。量子计算机使用量子位。量子位就像计算机中的常规位，只不过它有两种附加能力，即被置于叠加态和相互纠缠。
传统计算机执行确定性经典运算，也可以使用采样方法来模拟概率过程。通过利用叠加和纠缠，量子计算机可以执行难以用传统计算机大规模模拟的量子运算。利用 NISQ 量子计算的构想包括优化、量子模拟、密码学和机器学习。
量子机器学习
量子机器学习 (QML) 基于两个概念构建：量子数据和混合量子经典模型。
量子数据
量子数据是在自然或人工量子系统中出现的任何数据源。这可以是由量子计算机生成的数据，例如从用于证明 Google 的量子霸权的 Sycamore 处理器收集的样本。量子数据表现出叠加态和纠缠态，最终产生可能需要数量以指数级增长的经典计算资源来表示或存储的联合概率分布。量子霸权实验表明，可以从 2^53 个希尔伯特空间的极端复杂联合概率分布中进行采样。
NISQ 处理器生成的量子数据是嘈杂数据，而且通常在测量之前就发生纠缠。启发式机器学习技术可以创建最大程度地从嘈杂纠缠数据中提取有用经典信息的模型。TensorFlow Quantum (TFQ) 库提供了用于开发模型的基元，这类模型可以解开并归纳量子数据中的相关性，从而为改进现有量子算法或发现新的量子算法创造机会。
下面给出了可以在量子设备上生成或模拟的量子数据示例：
- 化学模拟 - 提取有关化学结构和动力学的信息，并将其潜在地应用于材料科学、计算化学、计算生物学和药物发现等领域。
- 量子物质模拟 - 对高温超导或表现出多体量子效应的其他奇特物质状态进行建模和设计。
- 量子控制 - 可以对混合量子经典模型进行变分训练，以执行最佳的开环或闭环控制、校准和误差抑制。这包括用于量子设备和量子处理器的错误检测与纠正策略。
- 量子通信网络 - 使用机器学习来区分非正交量子态，并将其应用于结构化量子中继器、量子接收器和纯化装置的设计与构造。
- 量子计量 - 量子增强的高精度测量（例如量子传感和量子成像）本质上是在探针这种小型量子设备上完成的，可以通过变分量子模型来设计或改进。
混合量子经典模型
量子模型可以表示和归纳包含量子力学起源的数据。由于近期的量子处理器仍然很小且嘈杂，因此量子模型无法仅使用量子处理器来归纳量子数据。NISQ 处理器必须与传统的协处理器协同工作才能生效。由于 TensorFlow 已经支持跨 CPU、GPU 和 TPU 的异构计算，因此被用作试验混合量子经典算法的基础平台。
量子神经网络 (QNN) 用于描述最好在量子计算机上执行的参数化量子计算模型。此术语通常可与参数化量子电路 (PQC) 互换。
研究
在 NISQ 时代，尚且无法在有意义的规模上实现比经典算法（例如 Shor 的分解质因数算法或 Grover 的搜索算法）更快的量子算法。
TensorFlow Quantum 的目标是帮助发现 NISQ 时代的算法，特别关注以下方面：
- 使用经典机器学习来增强 NISQ 算法。希望来自于经典机器学习的技术可以增强我们对量子计算的理解。在通过经典循环神经网络进行量子神经网络的元学习中，循环神经网络 (RNN) 用于发现对 QAOA 和 VQE 等算法的控制参数进行优化比简单的现成优化器更加有效。而用于量子控制的机器学习则使用强化学习来帮助减少误差并产生质量更高的量子门。
- 使用量子电路对量子数据进行建模。如果您有数据源的精确描述，则可使用经典方式对量子数据进行建模，但有时无法实现。要解决此问题，您可以尝试在量子计算机上建模并测量/观测重要的统计数据。量子卷积神经网络给出了一种量子电路，这种电路采用类似于卷积神经网络 (CNN) 的结构设计，可以检测物质的不同拓扑相。量子计算机保存数据和模型。传统处理器只能从模型输出中看到测量样本，而无法看到数据本身。在 Robust entanglement renormalization on a noisy quantum computer 中，作者学习使用 DMERA 模型压缩有关量子多体系统的信息。
量子机器学习的其他关注领域包括：
如未另行说明，那么本页面中的内容已根据知识共享署名 4.0 许可获得了许可，并且代码示例已根据 Apache 2.0 许可获得了许可。有关详情，请参阅 Google 开发者网站政策。Java 是 Oracle 和/或其关联公司的注册商标。
最后更新时间 (UTC)：2022-09-21。
[null,null,["最后更新时间 (UTC)：2022-09-21。"],[],[],null,["# Quantum machine learning concepts\n\n\u003cbr /\u003e\n\nGoogle's\n[quantum beyond-classical experiment](https://ai.googleblog.com/2019/10/quantum-supremacy-using-programmable.html)\nused 53 *noisy* qubits to demonstrate it could perform a calculation\nin 200 seconds on a quantum computer that would take 10,000 years on the largest\nclassical computer using existing algorithms. This marks the beginning of the\n[Noisy Intermediate-Scale Quantum](https://quantum-journal.org/papers/q-2018-08-06-79/) (NISQ)\ncomputing era. In the coming years, quantum devices with tens-to-hundreds of\nnoisy qubits are expected to become a reality.\n\nQuantum computing\n-----------------\n\nQuantum computing relies on properties of quantum mechanics to compute problems\nthat would be out of reach for classical computers. A quantum computer uses\n*qubits* . Qubits are like regular bits in a computer, but with the added ability\nto be put into a *superposition* and share *entanglement* with one another.\n\nClassical computers perform deterministic classical operations or can emulate\nprobabilistic processes using sampling methods. By harnessing superposition and\nentanglement, quantum computers can perform quantum operations that are\ndifficult to emulate at scale with classical computers. Ideas for leveraging\nNISQ quantum computing include optimization, quantum simulation, cryptography,\nand machine learning.\n\nQuantum machine learning\n------------------------\n\n*Quantum machine learning* (QML) is built on two concepts: *quantum data* and\n*hybrid quantum-classical models*.\n\n### Quantum data\n\n*Quantum data* is any data source that occurs in a natural or artificial quantum\nsystem. This can be data generated by a quantum computer, like the samples\ngathered from the\n[Sycamore processor](https://www.nature.com/articles/s41586-019-1666-5)\nfor Google's demonstration of quantum supremacy. Quantum data exhibits\nsuperposition and entanglement, leading to joint probability distributions that\ncould require an exponential amount of classical computational resources to\nrepresent or store. The quantum supremacy experiment showed it is possible to\nsample from an extremely complex joint probability distribution of 2\\^53 Hilbert\nspace.\n\nThe quantum data generated by NISQ processors are noisy and typically entangled\njust before the measurement occurs. Heuristic machine learning techniques can\ncreate models that maximize extraction of useful classical information from\nnoisy entangled data. The TensorFlow Quantum (TFQ) library provides primitives\nto develop models that disentangle and generalize correlations in quantum\ndata---opening up opportunities to improve existing quantum algorithms or discover\nnew quantum algorithms.\n\nThe following are examples of quantum data that can be generated or simulated on\na quantum device:\n\n- *Chemical simulation* ---Extract information about chemical structures and dynamics with potential applications to material science, computational chemistry, computational biology, and drug discovery.\n- *Quantum matter simulation* ---Model and design high temperature superconductivity or other exotic states of matter which exhibits many-body quantum effects.\n- *Quantum control* ---Hybrid quantum-classical models can be variationally trained to perform optimal open or closed-loop control, calibration, and error mitigation. This includes error detection and correction strategies for quantum devices and quantum processors.\n- *Quantum communication networks* ---Use machine learning to discriminate among non-orthogonal quantum states, with application to design and construction of structured quantum repeaters, quantum receivers, and purification units.\n- *Quantum metrology* ---Quantum-enhanced high precision measurements such as quantum sensing and quantum imaging are inherently done on probes that are small-scale quantum devices and could be designed or improved by variational quantum models.\n\n### Hybrid quantum-classical models\n\nA quantum model can represent and generalize data with a quantum mechanical\norigin. Because near-term quantum processors are still fairly small and noisy,\nquantum models cannot generalize quantum data using quantum processors alone.\nNISQ processors must work in concert with classical co-processors to become\neffective. Since TensorFlow already supports heterogeneous computing across\nCPUs, GPUs, and TPUs, it is used as the base platform to experiment with hybrid\nquantum-classical algorithms.\n\nA *quantum neural network* (QNN) is used to describe a parameterized quantum\ncomputational model that is best executed on a quantum computer. This term is\noften interchangeable with *parameterized quantum circuit* (PQC).\n\nResearch\n--------\n\nDuring the NISQ-era, quantum algorithms with known speedups over classical\nalgorithms---like\n[Shor's factoring algorithm](https://arxiv.org/abs/quant-ph/9508027) or\n[Grover's search algorithm](https://arxiv.org/abs/quant-ph/9605043)---are\nnot yet possible at a meaningful scale.\n\nA goal of TensorFlow Quantum is to help discover algorithms for the NISQ-era,\nwith particular interest in:\n\n1. *Use classical machine learning to enhance NISQ algorithms.* The hope is that techniques from classical machine learning can enhance our understanding of quantum computing. In [meta-learning for quantum neural networks via classical recurrent neural networks](https://arxiv.org/abs/1907.05415), a recurrent neural network (RNN) is used to discover that optimization of the control parameters for algorithms like the QAOA and VQE are more efficient than simple off the shelf optimizers. And [machine learning for quantum control](https://www.nature.com/articles/s41534-019-0141-3) uses reinforcement learning to help mitigate errors and produce higher quality quantum gates.\n2. *Model quantum data with quantum circuits.* Classically modeling quantum data is possible if you have an exact description of the datasource---but sometimes this isn't possible. To solve this problem, you can try modeling on the quantum computer itself and measure/observe the important statistics. [Quantum convolutional neural networks](https://www.nature.com/articles/s41567-019-0648-8) shows a quantum circuit designed with a structure analogous to a convolutional neural network (CNN) to detect different topological phases of matter. The quantum computer holds the data and the model. The classical processor sees only measurement samples from the model output and never the data itself. In [Robust entanglement renormalization on a noisy quantum computer](https://arxiv.org/abs/1711.07500), the authors learn to compress information about quantum many-body systems using a DMERA model.\n\nOther areas of interest in quantum machine learning include:\n\n- Modeling purely classical data on quantum computers.\n- Quantum-inspired classical algorithms.\n- [Supervised learning with quantum classifiers](https://arxiv.org/abs/1810.03787).\n- Adaptive layer-wise learning for quantum neural network.\n- [Quantum dynamics learning](https://arxiv.org/abs/1909.12264).\n- [Generative modeling of mixed quantum states](https://arxiv.org/abs/1910.02071) .\n- [Classification with quantum neural networks on near term processors](https://arxiv.org/abs/1802.06002)."]]
