What is machine learning?
This paper is for AI Xiaobai.
The author thinks that learning can be shown in three aspects: one is to speak out in public. Two, it can be written down; Third, it can be done. Of course, not everyone has the opportunity to speak in public, so write it down in your own words. How to express succinctly and correctly in your own language is a process of constant practice. Hope can insist.
Definition of machine learning
There is no widely accepted definition to describe machine learning. The first definition of machine learning comes from Arthur Samuel. He defined machine learning as the field of giving computers the ability to learn in the context of specific programming.
There are several different types of learning algorithms. The two main types are what we call supervised and unsupervised learning. In a nutshell, the idea of supervised learning is that we will teach the computer how to do a task (in the case of judging whether a person in a photograph is beautiful, we will tell the computer what beauty is when we enter a photograph. What is ugly; What’s right, what’s wrong…) In unsupervised learning, we want it to learn by itself (for example, if you feed a lot of data about pictures of animals, the computer will automatically classify them into different categories based on their various characteristics, which we don’t know). These two methods are described in detail below.
Supervised learning
Let’s take an example. As an interesting example, let’s say we’re going to implement a system to determine if there’s a kiss in a movie. So what we’re going to do with the system is — we’re going to put in a video, and we’re going to give you a specific time frame for kissing. So we use supervised learning to train the model of the system, first we need a lot of kissing, like kissing… Photo collection. And mark kissing as 1 and non-kissing as 0, and input a large number of labeled pictures into our pre-designed program. (This process is the training process, after the training, we will get a model, the model is the computer to learn the kissing result), then we use this model, using a lot of film and video test, each kiss picture will have a probability output. According to the probability can determine whether the time occurs.
Unsupervised learning
In supervised learning, every piece of training data is labeled yes or no, so for every piece of data in supervised learning, we already know the correct answer for the training set.
In unsupervised learning, we have data. It looks a little bit different, different from what supervised learning data looks like, which is unsupervised learning doesn’t have any labels or it has the same labels or it doesn’t have labels. So we know the data set, but we don’t know what to do with it, and we don’t tell each data point what it is. Nothing else. It’s just a data set. Can you find some kind of structure in the data? For a data set, unsupervised learning can determine that there are two different clusters of data. This is one, that is the other, they are different. Yes, unsupervised learning algorithms might divide the data into two different clusters. So it’s called a clustering algorithm. Examples of clustering algorithms, such as news categorization, automatically group similar stories into a category in the middle of a pile of stories and then display them.
This is just an overview of the two types of algorithms most commonly used in machine learning. There’s semi-supervised learning, where in a data set, you label some parts, and you don’t label some parts.