（四）multi-layer perceptron network
Can solve Exclusive Linear-non-classifiable input patterns，这解决了单层感知器的缺点.本质上是由于加了隐藏层的原因。
Each layer equivalent to a Single-Layer Perceptron Networksqth layer forms a nq-1 dimension Super-Plane perceptron networks, which can linearly classify the input patterns of this layer .Through Multi-Layer combination, eventually can implement the complex classification to input patterns.
下面是一个具体解决线性不可分思路。
最后总结下perceptron的优缺点
Advantages:Consists of Linear Threshold unitsIts Structure & Learning Algorithm is the basis of other Feed-forward NetworksAfter defining the Structure of Perceptron Networks:The Weights from input to hide layer can be obtained through StudyThe Weights from hide to output layer can use AND/OR method
Disadvantages:Tolerance & Accuracy of classification & Degree of approximation approach is still low compare with BP NN
