Neural Networks and Deep Learning-引论

什么是神经网络？

举个例子，假如用房子的size去预测价格，如下图：

很容易联想到了linear regression，但是如果这样看：

输入X:size后，经过这个neural(神经元)后，获得一个值y，那么这个circle就是一个function，function的功能可以被看做是linear regression的功能，那么可以理解成，输入x后，通过神经元neural，获得一个prediction y值，这就是最简单的神经网络。叫做 single neural network（单神经网络）。

if you stack many of the single neural network together, you will get a bigger and more complicated neural network.

So, let's see another figure below:

 

You want to predict the price of the house which is depended by many features. The X represents the features such as size, bedrooms and others which can influence the house price. The middle circle represents the functions(hidden factor) which takes their inputs from the four features. 

For example, x size and x bedrooms may decided the family size (first circle). Zip code decide the step (second circle). So, there is a hidden layer (隐藏层)。Then, after the hidden layer operations, we can get the outputs.

The neutral network is when the features are acted as the inputs, it can give us a y output as the prediction results.

                                                         Input  ------->hidden layer -------->output   

Why neural network becomes popular recently?

Because traditional machine learning algorithms can not handle big data while there are a huge data accumulated which is more than them were able to effectively take advantage of. 

Figure below shows the ability of traditional machine learning vs neural network.

When the amount of data is not huge, may be traditional algorithm is more effectively than neural network. But when facing with the huge amount of data, the neural network has its own advantages.

Two things you should notice:

1. Big enough neural network.

2. Huge amount of labeled data.