激光与光电子学进展, 2017, 54 (8): 080005, 网络出版: 2017-08-02
储备池计算硬件实现方案研究进展 下载: 1651次
Research Progress in Hardware Implementations of Reservoir Computing
光电子学 光信息处理 递归神经网络 储备池计算 非线性动力学系统 掩模 optoelectronics photonic information processing recurse neural network reservoir computing nonlinear dynamic system mask
摘要
储备池计算是一种适合处理时序信号的简单高效的机器学习算法。相比在传统电子计算机上用软件实现的方式, 储备池计算在光器件上的实现方式将更有利于超高速和超低功耗的信息处理。介绍了储备池计算的基本原理, 从输入层、储备池和输出层三个方面介绍了储备池计算硬件实现方案的研究进展, 指出了储备池计算硬件实现方案发展中存在的问题, 并展望了其未来发展趋势。
Abstract
Reservoir computing is a simple and effective machine learning algorithm to process time dependent signals. Compared with the software implementation in traditional electronic computer, reservoir computing implementation with optical components is more beneficial to information processing with ultrafast speed and ultralow power consumption. The basic principles of reservoir computing are presented, and the research progress in hardware implementation of reservoir computers is introduced from three aspects of input layer, reservoir and output layer. The existing problems in the development of the hardware implementation are demonstrated, and their future developing trends are discussed as well.
李磊, 方捻, 王陆唐, 黄肇明. 储备池计算硬件实现方案研究进展[J]. 激光与光电子学进展, 2017, 54(8): 080005. Li Lei, Fang Nian, Wang Lutang, Huang Zhaoming. Research Progress in Hardware Implementations of Reservoir Computing[J]. Laser & Optoelectronics Progress, 2017, 54(8): 080005.