Machine learning 8-artificial neural network

One . Artificial neural network

Artificial neural networks can do anything

1.1 The principles of the human nervous system

Simplified schematic diagram :

1.2 Learning and memory

Part of the structure is innate ( Unconditioned reflection )
Most of them are connected through learning ( It is also possible to eliminate the connection )
The job of memory is to change synaptic strength

1.3 The origin of artificial neural networks

1.4 Reference books

Two . ANN

2.1 ANN brief introduction

A term is used to explain :
Input node : x1,x2,x3
Output node : y
Weight vector : 0.3,0.3,0.3 （ The weight of each input node can be different ）
Bias factor : 0.4 （y=0.3*x1 + 0.3*x2 + 0.3*x3 - 0.4 ）
Activation function : For solving nonlinear problems
Learning rate :

2.2 Example

y Is the value of a symbolic function , The result of a symbolic function is usually 1 and -1, It can be understood as yes and no, By judging the input conditions , Then add the weight , The final judgment is yes still no.

2.3 Self learning algorithm

Description of algorithm ：

Convergence condition ：

Geometric interpretation :

The choice of learning rate :

Limitations of single-layer perceptron :

2.4 Training bias

2.5 Various activation functions

y=0.3*x1 + 0.3*x2 + 0.3*x3 - 0.4
The above functions deal with linear relations , But if the input value is nonlinear to the result , Then it will not be used at this time , The activation function came into being , To solve such problems .

2.6 Neural network topology

2.7 Learning rule

2.8 Artificial neural network modeling

2.9 Linear neural networks

Reference resources :

1. http://www.dataguru.cn/article-4063-1.html
2. https://zhuanlan.zhihu.com/p/25279356
3. https://zhuanlan.zhihu.com/p/98863801