| 摘要: |
| 提出一种能同时模拟包括M-P与TAF神经元在内的各种神经元通用的新数学模型,给出这种模型的一般形式.该模型不但连接权值可调节,而且激活函数可根据实际问题动态地选取。激活函数加入参数后,即成为变参数激活函数,大大地增强神经网络的灵活性. |
| 关键词: 神经元数学模型 激活函数 变参数 XOR问题 |
| DOI： |
| 投稿时间：2003-08-20 |
| 基金项目:广西自然科学基金资助项目[桂科基0141034]和广西高校百名中青年学科带头人资助项目。 |
|
| Variable Tunable Activation Function Neuron Mathematical Models |
|
Dai Zhenjie1, Zhou Yongquan2
|
| (1.Dept. of Math. & Comp., Guangxi College of Education, Jianzhenglu, Nanning, Guangxi, 530023, China;2.Dept. of Comp. & Info. Science Coll., Guangxi Univ. for Nationalities, 80 Daxuelu, Nanning, Guangxi, 530006, China) |
| Abstract: |
| A new kind of variable tunable activation function neuron mathematical model is designed,and the general form of variable tunable activation function is given.This form is different from M-P model,and the link weights is tunable.Since the variable tunable activation functions can be choosed randomly,so the non-linear capacity of neural networks could be improved.Finally,several given XOR problem examples show that the proposed new neuron mathematical model is effective and practical. |
| Key words: neuron model activation function variable XOR problem |
