A Novel Hardware-Oriented Recurrent Network of Asynchronous CA Neurons for a Neural Integrator
A Novel Hardware-Oriented Recurrent Network of Asynchronous CA Neurons for a Neural Integrator
复制标题
用于神经积分器的新型面向硬件的异步 CA 神经元循环网络
DOI：
10.1109/tcsii.2021.3063932
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Torikai Hiroyuki
中科院分区：
文献类型：
--
作者：
Takeda Kentaro;Torikai Hiroyuki
在这个简短的，递归网络推导出的神经积分器（NI）描述的常微分方程（ODE），可以用更少的计算资源来解决。随后，基于简化的ODE模型，提出了一种面向硬件的递归网络模型，用于由异步元胞自动机（CA）神经元组成的NI。所提出的模型，减少ODE模型，和以前提出的NI模型上实现的现场可编程门阵列。分析结果表明，所提出的模型消耗更少的硬件资源和更低的功耗比减少ODE模型和先前提出的NI模型。
In this brief, a recurrent network was derived for a neural integrator (NI) described by an ordinary differential equation (ODE) that can be solved with fewer computational resources. Subsequently, based on the reduced ODE model, a hardware-oriented recurrent network model was proposed for a NI consisting of asynchronous cellular automaton (CA) neurons. The proposed model, reduced ODE model, and the previously proposed NI model were implemented on a field-programmable gate array. Analysis results suggest that the proposed model consumes fewer hardware resources and lower power than those of the reduced ODE model and the previously proposed NI model.