量子纠缠:从量子物质态到深度学习  

Quantum entanglement: from quantum states of matter to deep learning

作　　者：程嵩[1,2] 陈靖[1,2] 王磊[1] CHENG Song CHEN Jing WANG Lei(Institute of Physics,Chinese Academy of Sciences, Beijing 100190,China University of Chinese Academy of Sciences, Beijing 100049,China)

机构地区：[1]中国科学院物理研究所,北京100190 [2]中国科学院大学,北京100049

出　　处：《物理》2017年第7期416-423,共8页Physics

摘　　要：量子纠缠在量子物质态的研究中扮演着日趋重要的角色,它可以标记传统范式难以区分的新奇量子态和量子相变,并指导设计高效的数值算法来精确地研究量子多体问题。最近,随着一些深度学习技术在量子物理问题中的应用,人们惊奇地发现:从量子纠缠的视角审视深度学习,或许有助于反过来理解和解决一些深度学习中的问题。量子纠缠定量化地刻画了现实数据集的复杂度,并指导相应的人工神经网络结构设计。沿着这个思路,物理学家们对于量子多体问题所形成的种种洞察和理论可以以一种意想不到的方式应用在现实世界中。Quantum entanglement is playing an increasingly significant role in the studies of quantum states of matter. It identifies novel phases and phase transitions beyond the traditional paradigms, and guides efficient simulation of quantum systems using classical computers. Recently, along with the application of deep learning technology to quantum many-body systems, a new perspective on deep learning emerges through the lens of quantum entanglement. Entanglement quantifies the complexity of a real dataset in machine learning and can guide the architecture design of artificial neural networks. Along this line, insights and theories originally developed for quantum many-body systems may find unexpected applications in real-world problems.

关 键 词：量子纠缠 张量网络 人工神经网络 深度学习 

分 类 号：O413[理学—理论物理] TP18[理学—物理]