The Ising model : teaching an old problem new tricks
The Ising model : teaching an old problem new tricks
复制标题
伊辛模型：教老问题新技巧
DOI：
--
复制
发表时间：
2010
期刊：
影响因子：
--
通讯作者：
G. Rose
中科院分区：
文献类型：
--
作者：
Zhengbing Bian;Fabián A. Chudak;W. Macready;G. Rose
在本文中，我们研究了物理上实现量子退火的硬件在机器学习应用中的使用。我们将展示如何在零温度和有限温度操作模式下利用硬件。在零温度下，硬件被用作伊辛能量函数的启发式最小化器，在有限温度下，硬件允许从相应的玻尔兹曼分布中采样。我们依靠量子力学过程来更有效地执行这两项任务，而不是通过传统计算机上的软件模拟。我们展示了如何塑造Ising能量函数来解决一系列监督学习问题。最后，我们通过构建在几个合成和真实数据集上使用量子退火训练的学习算法来验证硬件的使用。我们证明了这种使用量子力学硬件学习的新方法可以为许多结构化监督学习问题提供显着的性能提升。
In this paper we investigate the use of hardware which physically realizes quantum annealing for machine learning applications. We show how to take advantage of the hardware in both zeroand finite-temperature modes of operation. At zero temperature the hardware is used as a heuristic minimizer of Ising energy functions, and at finite temperature the hardware allows for sampling from the corresponding Boltzmann distribution. We rely on quantum mechanical processes to perform both these tasks more efficiently than is possible through software simulation on classical computers. We show how Ising energy functions can be sculpted to solve a range of supervised learning problems. Finally, we validate the use of the hardware by constructing learning algorithms trained using quantum annealing on several synthetic and real data sets. We demonstrate that this novel approach to learning using quantum mechanical hardware can provide significant performance gains for a number of structured supervised learning problems.