Optoelectronic Reservoir Computing Using a Mixed Digital-Analog Hardware Implementation
Optoelectronic Reservoir Computing Using a Mixed Digital-Analog Hardware Implementation
复制标题
使用混合数模硬件实现的光电储层计算
DOI：
10.1007/978-3-030-30493-5_18
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Ingo Fischer
中科院分区：
文献类型：
--
作者：
M. C. Soriano;P. Massuti;J. Yelo;Ingo Fischer
光电系统已被证明是实现基于硬件的油藏计算（RC）的一个有吸引力的平台。这些非常规计算机可以在训练后执行混沌时间序列的非线性预测并生成任意输入/输出函数。基于延迟的光电储层的主要优点之一是它只需要单个非线性硬件节点和反馈回路。为了通过实验实现 RC 方案，我们使用光子和电子元件以及 FPGA 来驱动系统。由此产生的光电 RC 是一个具有强大多功能性的混合模拟和数字系统。我们表明，这种设置可以执行混沌时间序列预测任务，其中输出是在线计算的，预测误差较低。最终，该测试台系统将允许测试多种策略以提高预测性能，包括添加多个延迟和定制输入掩模以减轻噪声。
Optoelectronic systems have proven to be an attractive platform for the realization of hardware-based Reservoir Computing (RC). These unconventional computers can perform nonlinear prediction of chaotic timeseries and generate arbitrary input/output functions, after training. One of the main advantages of a delay-based optoelectronic reservoir is that it only requires a single nonlinear hardware node and a feedback loop. To implement the RC scheme experimentally, we use photonic and electronic components and an FPGA to drive the system. The resulting optoelectronic RC is a hybrid analog and digital system with great versatility. We show that this set-up can perform a chaotic timeseries prediction task, in which the output is computed online, with low prediction errors. Ultimately, this testbed system will allow to test several strategies to improve the prediction performance, including the addition of multiple delays and the tailoring of the input mask for noise mitigation.