(4) Neural Networks: Representation

The following content is derived from machine learning on Coursera and is based on Rachel-Zhang's blog (http://blog.csdn.net/abcjennifer)

After talking about the two common methods of logisitc regression and linear regression, we need to learn more about other machine learning methods considering some disadvantages,

Abstract:

(1) (2): it helps us understand some basic concepts of neural networks;

(3) (4) neuronsand
The brain I & II: Working Principle and calculation formula of each layer of the neural network. (the most important part in this section)

(5) (6) examplesand
Intuitions I
& II: How does a neural network perform complex logical operations, that is, complex hypothesis?

(7)

Multiclassclassification:

What we mentioned above is for the second class. Should we do this for multiple classes?

Body:

**************************************** **************************************** **************************************** ***************

(1) nonlinear hypothesis:

The linearregression and logistic regression mentioned above are aimed at a small number of features. However, what if the number of features is large? If you still use the above two methods, you will not be able to cope with the problem, because not only is the operation slow, but it is also possible to overfitting.

The boundarydesion in this image is a pink curve, which may cause overfitting if logistic regression is used.

 

In short, in some cases, the two methods mentioned above are no longer applicable to classification. So what does it use? Answer: Neural Networks

**************************************** **************************************** **************************************** ***************

(2) neuronsand the brain

This section describes the evolution and decline of neural networks, and the rise of neural networks. Some amazing images and applications are provided. If you are interested, you can check these interesting things online.

(Because there is no substantive content, it is to give a concept, so there is no .)

**************************************** **************************************** **************************************** ***************

(3) neuronsand the brain I & (4) neurons and the brain II

Here we will not talk about the similarities between neural networks and our human brain. This article describes how a neural network works.

 

1. Give the simplest logistic uint:

(There are only two layers. The first layer is the x0, x1, x2, X3, which is called the input layer. The second layer is the brown circle, which is called the output layer)

2. The following shows the most basic layer-3 model, and shows how to do it layer by layer.

(Note that the following two images are important at the beginning of the neural network. They illustrate the specific working principles! If you do not understand it, read it again until you understand it !!)

We will not talk about how formulas are made. First, let's take a look at the upper and lower coordinates and their meanings.

(1) first, remember that the coordinates of any letter represent the layers! (If it is 1, it indicates the input layer, and 2 indicates the hidden layer ).

(2) The coordinate below represents the number of a layer.

(3). It is a matrix (the size of the matrix is determined by the number of layers J and J + 1 !). Careful colleagues found that the coordinate is not only a number, but also composed of two numbers. In fact, these numbers can be understood as the abscissa and ordinate in the array, but here the abscissa starts from 1 and the ordinate starts from 0 (why does it start from 0? The figure below will be discussed)

After talking about the meanings of the three elements, we should analyze how the formula is launched.

Let's first explain the above-mentioned questions: why does the ordinate start from 0?

The figure in the upper left corner is essentially the same as the figure in the upper left corner, except that 1 is added to the top of each layer, this is like logistic regression, which adds the same principle of 1 before each feature. In fact, this is why I mentioned above that the ordinate starts from 0.

**************************************** **************************************** **************************************** ***************

Knowing how each layer works, let's take a look at why neural networks can handle complex and non-linear hypothesis.

(5) examplesand intuitions I (6) examples and intuitions II

Everyone knows or, and, not operations.

 

**************************************** **************************************** **************************************** ***************