A Novel Generalized Hardware-Efficient Neuron Model Based on Asynchronous CA Dynamics and Its Biologically Plausible On-FPGA Learnings
A Novel Generalized Hardware-Efficient Neuron Model Based on Asynchronous CA Dynamics and Its Biologically Plausible On-FPGA Learnings
复制标题
DOI：
10.1109/tcsii.2018.2876974
复制
发表时间：
2019-07
期刊：
影响因子：
--
通讯作者：
Taiki Naka;H. Torikai
中科院分区：
文献类型：
--
作者：
Taiki Naka;H. Torikai
本文提出了一种新的广义神经元模型(即胞体-树突-棘模型)，其动力学由一个异步元胞自动机来描述。结果表明，神经元模型及其网络模型可以实现三种典型的生物似然学习。这些模型在现场可编程门阵列(FPGA)中实现，实验验证了学习的操作。结果表明，与传统模型相比，提出的模型消耗更少的硬件资源(例如，不到37%的FPGA片)和更低的功耗(例如，不到39%的功耗)。
This brief presents a novel generalized neuron model (i.e., soma-dendrite-spine model) the dynamics of which is described by an asynchronous cellular automaton. It is shown that the neuron model and its network models can realize three kinds of typical biologically plausible learnings. These models are implemented in a field programmable gate array (FPGA) and experiments validate operations of the learnings. It is then shown that the presented models consume fewer hardware resources (e.g., less than 37% FPGA slices) and lower power (e.g., less than 39% power) compared to conventional models.