The article reads the relationship between artificial intelligence, machine learning, neural network and deep learning

Some time ago I saw a lot of book blogs and forums on artificial intelligence. I deeply felt that artificial intelligence is a pit. There are too many knowledge points and disciplines in it. If you want to dig deeper than it is easy, then you will write some of your own. The blogger's idea of ​​recording his own learning history and some knowledge points, gives him a summary of the harvest, and at the same time supervises his own motives. This is also a "supervised study":)

Here is referred to as "supervised learning", which is often seen when we first started learning AI/Machine Learning. It takes a little time for the concept to learn from ignorance and ignorance. These concepts are so basic that they don't have a lot of space to introduce them. However, for newcomers, when they first come into contact with a new field, they often see a lot of nouns called “tall”. The learning curve is steep, so let's start with the basic concepts. Therefore, this note is a carding of the basic concepts. If there is something wrong, we cannot wait to enlighten me.

Artificial intelligence, machine learning, neural networks, deep learning relationships

When I first contacted AI content, I often saw artificial terms, machine learning, deep learning, and different neural network terms. Each one was so cold that I couldn’t distinguish between them. What kind of relationship, in many cases, is considered to be a different expression of a thing. After seeing some specific introductions, it gradually became a general model.

Machine learning

Machine learning is the most important content of artificial intelligence. Let's take a look at its definition (of course, there are many different definitions): "Machine learning is the idea that there are are algorithms that that can tell you something interesting about a set of data without You having to write any custom code specific to the problem. Instead of writing code, you feed data to the generic algorithm and it builds its own logic based on the data.” There are a few important keywords that you don’t need to write Instead of specialized business logic code, by inputting a large amount of data to the machine, the machine establishes its own business logic through a general mechanism, that is, the machine "self-learns" the logic of the business. Of course, this learning logic can be Used to process new data. This is similar to the human learning process, as shown below:

Supervised learning and unsupervised learning

These two concepts are also concepts that have just come into contact with machine learning. Popular/simple point, so-called supervised learning is that training history data has both problems and answers, while unsupervised learning is training historical data only. There is no answer to the question. The formal statement is generally called the label label and there is a mixed learning method between the two, called semi-supervised learning.

In unsupervised learning, it is mainly the discovery of unknown structures or trends in the data. Although the original data does not contain any tags, we hope that we can integrate (group or cluster) the data or simplify the data (dimensionality reduction, removal of unnecessary variables, or detection of outliers). Therefore, the main categories of unsupervised algorithms include: - Clustering algorithm (representative: K-means clustering, system clustering) - Dimensionality reduction algorithm (represented: PCA, LDA)

Supervised learning can be subdivided based on the type of predictor variable. If the predictors are continuous, then this is a regression problem. However, if the predictor is an independent category (qualitative or deterministic discrete), then this is a classification problem. The main categories of supervised learning therefore include: - Regression algorithms (linear regression, least-squares regression, LOESS local regression, neural network, deep learning) - classification algorithms (decision trees, support vector machines, Bayes, K-nearest neighbors Algorithm, Logistic Regression, Random Forest)

There are a lot of algorithms mentioned here. At present, there is no need to grasp them one by one. I believe that you will often see them in the future.

Of all these algorithms, the current hottest is probably deep learning, but to understand deep learning it is necessary to understand its predecessors (predecessors, fathers).

Neural Networks

There are many and many introductions to the neural network. There are many introductions and courses for the big cows. I mainly refer to/recommend the following: Neural network: The introduction of neural networks from neurons to deep learning is usually not described. In detail, an incomplete summary picture is made:

Well, the basic concept of a general machine learning is summarized. In fact, learning these basic concepts is still relatively simple and convenient. After all, we have a powerful search engine. As long as the input "machine learning" can get a lot of knowledge let us To learn, but for beginners to begin, you can try it out first, with a framework of understanding, to prepare for follow-up study.