However, "artificial intelligence" didn't happen alone, and he also had two inseparable teammates: "machine learning" and "deep learning". These three words appear in various places, like a combination of celestial groups, and sometimes even incarnate each other. Then the question is coming. What is the relationship between artificial intelligence, machine learning and deep learning? What are the connections and differences between them? We don't split the concept here, starting with the development of artificial intelligence.
The past and present of artificial intelligence
In the summer of p>1956, a group of young scientists, led by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon, gathered in Dartmouth to study and discuss a series of related problems of using machines to simulate intelligence (known as Dartmouth in history). At that meeting, the word "artificial intelligence" was put forward, which also marked the official birth of the emerging discipline "artificial intelligence".
At that time, the research of artificial intelligence was in the "reasoning period", and people thought that as long as machines were endowed with the ability of logical reasoning, they would have intelligence. At that time, the research did achieve some results, such as proving many mathematical theorems, and some theorems were even more ingenious than mathematicians.
However, human intelligence comes not only from logical reasoning ability, but also from a lot of experience and knowledge. For example, if I have never been on a plane or bought a plane ticket for others, when you ask me how much the plane ticket to Beijing will cost tomorrow, I guess I will be overwhelmed, and I don't even know Ctrip, so I am not so "intelligent", but in fact my reasoning ability has not dropped. Since the 197s, the research of artificial intelligence has entered a "knowledge period", and people hope to sum up the knowledge in various fields and tell it to the machine on the basis of reasoning, so that it can acquire intelligence. At that time, a large number of expert systems (program systems with a lot of expertise and experience, which can be used for reasoning and judging, and simulate the decision-making process of human experts) came out and made many achievements in many application fields.
But people soon realize that it is very difficult to sum up knowledge and teach it to machines (called "knowledge engineering bottleneck"), because there is so much knowledge of human beings that they have to write it into a form that machines can understand. If only we could give the machine some relatively primitive data, and then let the machine learn by itself. So since 198s, the technical route of machine learning has gradually dominated the research of artificial intelligence until now.
What is machine learning
Machine learning can be understood as that the machine searches for and refines (trains/learns) some rules (models) from the known empirical data (samples) through a certain method (algorithm); The extracted rules can be used to judge some unknown things (prediction).
for example, if we contact 1w single Wang (known sample), we can find out some * * * sexual characteristics of TA through induction, summary and comparison (algorithm), and then use these * * sexual characteristics as the basis (model) for judging single Wang, so we can judge whether TA is single (prediction) next time we meet a person (unknown sample).
since we are looking for the rule from a bunch of known samples, the way to find the rule and the shape of the rule will vary from person to person, that is, the algorithm and model may be different. Therefore, machine learning itself is divided into different schools, and each school has its representative model and algorithm. Machine learning is mainly divided into symbolic learning (represented by decision tree model and related algorithms), connectionist learning (represented by neural network model and related algorithms) and statistical learning (represented by support vector machine and related algorithms). Symbolism learning and connectionism learning were very popular from the 198s to the mid-199s, while statistical learning quickly occupied the stage from the mid-199s. It is worth mentioning that a series of characteristic laws found out to judge whether a person is single or not are actually a decision tree.
The rise of deep learning
After entering the 21st century, the rise of Internet and mobile Internet has caused the explosive growth of data, cloud computing has also greatly enhanced the computing power, and at the same time, the related algorithms of neural networks have gradually matured, which has led to the resurgence of neural networks, the representative of Connectionism. The neural network that comes back again often has a larger network hierarchy than before, so it is called "deep neural network". With sufficient training data and computing power, deep neural network has achieved excellent performance in many complicated tasks, especially speech processing, natural language processing and image processing. The breakthrough in performance has promoted the application of artificial intelligence in a series of scenes such as speech recognition, text translation and face recognition, which has made everyone see the economic benefits and imagination space brought by the landing of new technologies, thus causing the upsurge of artificial intelligence.
Summary
Having said that, let's use a picture to illustrate the technical genre category and evolution of artificial intelligence. It should be noted that although the genre of artificial intelligence is constantly evolving, this does not mean that the past technical route has been abandoned. This is more like a practical application-oriented style-in an era, a certain technical genre can just solve the actual industrial problems that need to be solved in this era, so it will naturally become popular. At present, many different machine learning technologies are being applied to their own suitable scenarios, such as support vector machine, which is the representative of statistical learning, and is still the first choice for text classification tasks.
Finally, answer the questions in the title of the article. Artificial intelligence is a big concept, and it is a subject that studies how to make machines acquire intelligence. Machine learning is a technical school in artificial intelligence, which obtains the "intelligence" of judging unknown samples by extracting rules from known samples; Deep learning is a kind of machine learning, and the model it learns is deep neural network.