基于脉冲神经网络的模式识别模型研究

论文总字数：40952字

摘 要

关键词 …………………………………………………………………………………………Ⅰ

Abstrac…………………………………………………………………………………………Ⅱ

Key words………………………………………………………………………………………Ⅱ

1绪论……………………………………………………………………………………………1

1.1脉冲神经网络的提出…………………………………………………………………1

1.2背景及发展意义………………………………………………………………………1

1.3脉冲神经网络原理介绍………………………………………………………………2

1.4实证检验方法述评……………………………………………………………………3

1.5本文的结构安排………………………………………………………………………3

2 国内外脉冲神经网络文献研究………………………………………………………………4

2.1 国外研究成果综述……………………………………………………………………4

2.1.1 硬件部分综述…………………………………………………………………4

2.1.2 软件部分综述…………………………………………………………………8

2.2国内研究现状及现存问题……………………………………………………………15

2.3小结……………………………………………………………………………………16

3脉冲神经网络监督学习算法的基本框架……………………………………………………18

3.1脉冲神经网络监督学习算法的基础构成……………………………………………18

3.2脉冲神经网络监督学习算法相关性能评价…………………………………………19

4脉冲神经网络监督学习算法分析……………………………………………………………20

4.1突触可塑性学习规则…………………………………………………………………20

4.1.1突触可塑性学习算法的背景基景……………………………………………20

4.1.2突触可塑性监督学习算法的分类……………………………………………21

4.1.3突触可塑性监督学习算法的性能比较………………………………………23

5实例仿真与比较分析…………………………………………………………………………24

5.1常用的脉冲神经网络模型……………………………………………………………24

5.2模拟与仿真……………………………………………………………………………28

6结论和展望……………………………………………………………………………………34

参考文献………………………………………………………………………………………35致 谢……………………………………………………………………………………………37

摘 要

脉冲神经网络（SNN-Spiking Neuron Networks) 经常被誉为第三代人工神经网络。第一代神经网络被称为感知器，它仅仅是一个非常单一的神经元模型，并且最多只能对二进制数据进行处理。第二代神经网络的包含范围相对而言更为扩大，例如BP神经网络。凭借针对神经学术的研究与努力，研究学者已经能够构建较为精确的、建立在脉冲时间产生之上的神经网络模型。这种神经网络相较之前传统的神经网络更为新颖，使用脉冲编码(spike coding)，通过精准获取脉冲发生的时间，能够获得更多的信息，其计算能力也得到进一步提升。传统人工神经网络虽然在许多领域得到了成功应用，但随着研究的深入，由于其没有考虑时间编码而带来的局限性，存在的问题正逐渐显现。因此，脉冲神经网络应运而生，且由于其优越的性能和显著增强的计算能力，获得了诸多研究人员的关注。

但是，因为脉冲神经网络的复杂程度相对较高，到目前研究者仍然没能找到较为高效的算法。本课题将对比当前主流算法及技术，分析并对其中最为有效的、适合于多层次多空间的、脉冲神经网络的通用学习算法及模型进行仿真。

关键词：脉冲神经网络，人工神经网络，监督学习算法，模拟，仿真

Abstract

SNN-Spiking Neuron Networks is often known as the third generation of artificial neural networks. The first generation of neural networks is called a perceptron, it is only a very single neuron model, and can only deal with binary data. The inclusion range of the second generation neural network is relatively larger, such as BP neural network. With the study and effort of neuroscience, the research scholars have been able to construct a more accurate neural network model based on the pulse time. This kind of neural network is more novel than the traditional neural network, using the spike coding (spike coding), through the accurate acquisition of the pulse time, can get more information, its computing power has been further improved. Although the traditional artificial neural network has been successfully applied in many fields, with the deepening of the research, because of its lack of consideration of the limitations of time coding, the existing problems are gradually emerging. Therefore, the pulse neural network came into being, and because of its superior performance and significantly enhanced computing power, access to a lot of researchers attention.

However, because the complexity of the pulse neural network is relatively high, to the current researchers still can not find a more efficient algorithm.This subject will simulate and compare the current mainstream algorithms and techniques, and analyze and generalize the general learning algorithm of pulse neural network which is the most effective and multi-level multi-space.

KEY WORDS: Spiking neural network, artificial neural network, supervised learning algorithm,simulation, simulation

剩余内容已隐藏，请支付后下载全文，论文总字数：40952字