量子神经网络的随机性增强表达力
Yadong Wu1,2,3, Juan Yao4,5,6, Pengfei Zhang1,3
1Department of Physics, Fudan University, Shanghai 200438, China.
Physical review letters
|January 19, 2024
概括
研究人员通过向量子电路添加一个随机层来增强量子神经网络 (QNN). 这种新的方法改善了QNNs.
科学领域:
- 量子计算是一种量子计算.
- 人工智能的人工智能
- 机器学习 机器学习
背景情况:
- 量子神经网络 (QNN) 是人工智能和量子计算的混合体,对杂的量子设备显示出希望.
- 传统的QNN使用参数化量子电路进行操作和测量.
研究的目的:
- 为了增强QNN的表达力.
- 为提高性能,在量子电路中引入随机性.
主要方法:
- 整合了一种新的随机层,其中包含从可训练组合聚合中采集的单量子比特门.
- 利用乌尔曼定理进行大化来证明目标运算符的准确近似.
- 进行了数值实验,包括可观测学习,雷尼 Entropy 测量和图像识别.
主要成果:
- 通过引入随机性来证明QNN的增强表达性.
- 展示了近似任意目标运算符的能力,使可观测的学习成为可能.
- 在多个量子机器学习任务中验证了该方法.
结论:
- 提出的随机层显著提高了QNN表达力.
- 这种方法在量子机器学习中提供了广泛的应用.
- 随机性是提高各种任务的QNN性能的一个关键因素.
相关概念视频
01:37Neural Regulation
39.4K
Digestion begins with a cephalic phase that prepares the digestive system to receive food. When our brain processes visual or olfactory information about food, it triggers impulses in the cranial nerves innervating the salivary glands and stomach to prepare for food.
39.4K
01:23Atomic Nuclei: Nuclear Relaxation Processes
655
In the absence of an external magnetic field, nuclear spin states are degenerate and randomly oriented. When a magnetic field is applied, the spins begin to precess and orient themselves along (lower energy) or against (higher energy) the direction of the field. At equilibrium, a slight excess population of spins exists in the lower energy state. Because the direction of the magnetic field is fixed as the z-axis, the precessing magnetic moments are randomly oriented around the z-axis.
655