Machine Learning:Neural Network---Representation

时间:2023-03-09 20:21:10
Machine Learning:Neural Network---Representation

Machine Learning:Neural Network---Representation

1。Non-Linear Classification

假设还採取简单的线性分类手段。那么会面临着过拟合以及效率低下的问题(如图所看到的),然而neural network则能够非常好的解决非线性分类问题。

Machine Learning:Neural Network---Representation

watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvY3N0b3Bjb2Rlcg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="">

2,Model representation

Machine Learning:Neural Network---Representation

watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvY3N0b3Bjb2Rlcg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="">

第一层称为input layer,最后一层称为output layer,中间其余各层称为hidden layer。

Machine Learning:Neural Network---Representation

Machine Learning:Neural Network---Representation

注意一下权重參数theta的维数问题。

3。Forward propagationMachine Learning:Neural Network---Representation

watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvY3N0b3Bjb2Rlcg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="">

Machine Learning:Neural Network---Representation

Machine Learning:Neural Network---Representation

4。神经网络Example

Machine Learning:Neural Network---Representation

Machine Learning:Neural Network---Representation

watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvY3N0b3Bjb2Rlcg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="">

Machine Learning:Neural Network---Representation

watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvY3N0b3Bjb2Rlcg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="">

Machine Learning:Neural Network---Representation

Machine Learning:Neural Network---Representation

神经网络建模的难点在于神经结构的选择以及权重參数theta的选择,一个好的神经网络是须要非常精细的设计的。

5,Multi-class classification

Machine Learning:Neural Network---Representation

watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvY3N0b3Bjb2Rlcg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="">

Machine Learning:Neural Network---Representation

当将神经网络运用于Multi-class classification问题时。其输出h(theta)不再是一个数值而是一个向量,而且其值为1的元素相应着合适的分类。

----------------------------------------------------------------------------------------------------------------

本内容摘自斯坦福大学Andrew Ng老师《机器学习》课件