Neural network (神经网络)
A neural network is a massively parallel distributed processor made up of simple processing units that has a natural propensity for storing experiential knowledge and making it available for use. It resembles the brain in two respects:
- Knowledge is acquired by the network from its environment through a learning process.
- Interneuron connection strengths, known as synaptic weights, are used to store the acquired knowledge.
理解:神经网络模拟人的两个特点:
-
- 通过从环境获取信息进行学习;
-
- synaptic weights (突触权重)用于储存获得的信息。
Models of a neuron (神经元模型)
- 模型中有一系列的突触(synapses),也叫连结点(connection links),其特征是权重(weight)和强度(strength)。信号
j
通过突触j
,被w_kj
加重之后,传入神经元k
。 - 通过一个 adder 对输入信号进行求和。
- 通过一个 activation function 函数对输出结果的范围进行限制。例如常用的 sigmoid function。如果使用 McCulloch–Pitts 模型作为 activation function,神经元模型就变成随机模型(stochastic),带有嗓音(noise)
如果加入偏差值(bias),模型就变成:
信号流图(signal-flow graphs)
连结可以有不同的方式:
反馈机制
神经元模型中的输出结果y
可以把结果反馈给输入信号:
写成数学公式就是:
参考文献:
Neural Networks and Learning Machines by Simon Haykin