What is machine learning?

What is machine learning?

Not long ago, a good article, mainly about machine learning is what the subject or research field, but also introduced the relationship between ML and AI. The writer is a Zhou Zhihua teacher at Nanjing University. Talk less, directly on the article.

-------------------------------------------------------

Machine learning is now a big hit, with more people in the study, more and more newcomers pouring in.

Many people actually do not really think, this is not what they like to engage in things, just see others are engaged, feel with the people go always will not suffer.

The question is, is there really a "everybody"? Wouldn't it be "two boys" and "three boys"? If there are "a few gang", then in the end should follow the "What Gang" walk?

Many people may not realize that the so-called machine learning Community, now contains at least two groups with completely different cultures, completely different values, called machine learning "communities" may be more appropriate.

The first community is a group of machine learning as a branch of artificial intelligence , the subject of which is a computer scientist. Now "machine learning researchers" may have very few people who read the 1983 Learning:an Artificial Intelligence approach book. The publication of this book marks the beginning of machine learning as an independent field in artificial intelligence. It is actually a collection of early machine learning research, acquisition a number of sages (for example, Herbert Simon, the Nobel Prize, the Turing Award and a variety of other related awards almost all the scientific genius), editor is Ryszard S. Michalski (This gentleman has been dead for many years, he is one of the founders of machine learning), Jaime G. Carbonell (this June was Springer's Lnai editor), Tom Mitchell (This is the first department of the CMU Machine Learning Department director, the author of the famous textbook) , the machine learning Community has no one to know about him. The founding of Machine Learning Magazine is the result of the efforts of this group of people. The book is worth reading. Although technical means have changed rapidly, there are some profound ideas that are not outdated. There are always a lot of things in various disciplines, and after changing the new clothes, and now the transfer learning, in fact, is learning by analogy upgrade version.

The research of artificial intelligence from the focus of "reasoning" to "knowledge" as the focus, and then to "learning" as the focus, there is a natural, clear context (playing a now fiery analogy, from design features to learning characteristics, reprinted note). Ai-born machine learning researchers, most of them, use machine learning as a way to implement AI, as the title of the 1983-year book. They are concerned with the problem of artificial intelligence, hoping to use machine learning as a means, but specifically what kind of learning methods are based on statistical, algebraic, or logical, geometric , they are not care.

This group of people may not be satisfied with the current dominating status of statistical learning. Statistical learning is unlikely to solve most of the problems in AI, and if statistical learning suppresses the study of other means, it may not be a good thing. This group of people often does not care in the article to show their own mathematical level, and may even be to simplify the expression of their own ideas proud. the problem of artificial intelligence is not a mathematical problem, not even a problem that can be solved by mathematics. The difficulty of many things in artificial intelligence is that we do not know where the nature of the difficulty is, and where the "problem" is. Once the "problem" is clear, it may not be difficult to solve.

The second community is that machine learning is a group of "Applied Statistics" , the subject of which is statisticians.

Compared with pure mathematics, statistics are not very "clean", many mathematicians even refused to admit that statistics are mathematics. But if compared with artificial intelligence, statistics is too clean, the problem of statistical research is clear, not like artificial intelligence, even where the problem is not known anywhere. For quite a long time, statisticians and machine learning have remained at a distance. Slowly, many statisticians have come to realize that statistics are inherently application oriented, and machine learning is inherently a good entry point. Because machine learning uses a variety of mathematics, it is essential to analyze the laws contained in large amounts of data. The majority of the machine learning researchers who are statistically born are using machine learning as a statistical application. Their focus is on how to turn the theory and method of statistics into an algorithm that can be effectively implemented on a computer, which is useful for any problem in AI, and they are not care.

This group of people may have no interest in artificial intelligence, in their eyes, machine learning is statistical learning, is a branch of statistical comparative bias, at best, the intersection of statistics and computer science. This group of people are often excluded from the study of statistical learning, which is natural, based on algebraic, logical and geometrical learning, it is difficult to include the category of statistics.

The culture and values of the two groups are completely different. The first group thinks that a good job, the second group may feel that there is no technical content (even when reading the first group of papers, you will feel that this is "water", it is not much contribution. This indicates that you have been unconsciously affected by the second group. Reprinted by the note). But the first group may just think that simple is good, because it is good to grasp the nature of the problem, so the problem becomes easy to solve. The second group to appreciate the work, the first group may feel is a trick, see what he wants to solve any AI problem, is not engaged in artificial intelligence, computer, but others did not say that they are in the "artificial Intelligence", "engage in computer", is not in the artificial intelligence to do research.

Two groups have their own meaning of existence, should be tolerant of a little, do not need to go to each other than what. But since the machine learning this hat is not "a group of children", but "two boys", then to "follow up" the new people will be cautious, first make clear that they prefer "which gang son."

Cited two famous scholars to the end, one is the Artificial Intelligence award winner, a statistical study everyone, the name I do not say, save trouble:

"I do not come to AI to do statistics"

"I do not have the interest in AI"

What is machine learning?