基于固态单自旋体系的量子机器学习研究

In this era of big data, machine learning has been demonstrated to be an extremely powerful tool to tackle many problems such as finding patterns in data. Such a boost is mainly attributed to the increase of computing power and emergence of advanced algorithms. On the other hand, quantum machine learning, lying at the intersection of quantum physics and machine learning, promises us a substantial “quantum speedup” for many machine learning tasks, yet its experimental demonstration remains of great challenge..The nitrogen-vacancy (NV) center in diamond is one of the leading candidates for quantum information science thanks to its combination of solid-state host environment and functionalization even under ambient conditions. Exploring quantum machine learning algorithms on NV centers would not only enable first demonstrations of important quantum machine learning algorithms, but some of the algorithms can also be applied to assist in the design or optimization of coherent controls. In this project, we aim at the following goals: finding a hybrid solid system with four to six qubits where the electron, nitrogen and several carbon-13 spins can be addressed; realizing determination of Hamiltonian, efficient initialization and readout in that quantum register; transferring coherent control and feedback control techniques developed in nuclear magnetic resonance to the NV center to improve the control precision; implementing proof-of-principle quantum machine learning algorithms using four to six qubits, such as the quantum principal component analysis or quantum autoencoders. We hope this research will shed light on the understandings of experimental quantum machine learning, paving the way towards its realization in large scale.

借助于计算能力的提高和相关算法的发展，机器学习是当前处理大量数据问题的强有力工具。基于量子力学规则的量子机器学习可以为很多重要的机器学习任务提供“量子加速”，其实验研究是当前量子信息领域的研究重点和难点。.作为目前唯一的室温固态量子计算候选体系，基于金刚石色心的单自旋量子调控研究紧扣国家量子信息科学的战略布局。量子机器学习与金刚石色心的结合有双重意义：重要的量子机器学习算法将首次在室温固态体系演示；很多算法可用于辅助相干操控技术的设计和优化。本项目拟开展的研究内容包括：寻找电子自旋和邻近核自旋组成的复合体系，构建四至六个量子比特的寄存器；实现该体系的哈密顿量重构、有效初始化和读出等；基于申请人在核磁共振操控技术上逾十年的经验积累，将高比特控制技术移植至金刚石色心，提升控制精度；实验演示重要的量子机器学习算法，例如可用于图像识别的量子主成分分析或可用于量子态数据压缩的量子自编码器等。

随着计算能力的提高和算法的进步，机器学习已成为很多应用场景和领域不可或缺的研究工具。然而，随着模型复杂度的增加，经典计算机的算力和成本限制正成为机器学习的发展瓶颈。基于量子力学规则的量子机器学习可以为很多重要的机器学习任务提供“量子加速”，有望成为机器学习发展的新突破口。.量子机器学习的实验实现是该领域的难点之一。金刚石氮空位色心（NV色心）作为被广泛认可的室温量子计算候选体系，具有控制精度高、相干时间长、可扩展性好的特点。本项目以NV色心作为核心体系进行了量子机器学习的研究，研究内容分步展开：首先发展了自旋体系上的高保真度初始化、高精度相干操控和读取等技术，并完成了重要的量子计算和量子模拟实验演示，如伪随机数生成、动力学量子相变、三维手性拓扑绝缘体、不定因果序量子冰箱等；制备了高质量的NV色心样品，结合发展出的自旋调控技术，构建了包括电子自旋、氮自旋和近邻碳13自旋在内的4比特量子寄存器；理论上发展了用于量子态重构、哈密顿量测量、纠缠结构表征等的量子机器学习方法；实验上演示了量子主成分分析和量子自编码器等量子机器学习任务，成功实现了人脸图像的初步识别和NV色心电子-核自旋纠缠态寿命的延长等。.在项目支持下，总计发表含致谢本项目的文章28篇，包括项目负责人以通讯作者身份发表了PRL论文6篇和npj Quantum Inf. 论文4篇。团队负责人鲁大为获批国家青年特聘专家（2019年）、深圳市鹏城学者特聘教授（2020年），团队成员李俊获批国家优秀青年科学基金（2022年）、广东省杰出青年基金（2021年），团队成员辛涛获批广东省杰出青年基金（2022年）等多项人才项目。