Machine learning and artificial Intelligence Learning Resource guidance

Machine learning and artificial Intelligence Learning Resource guidance

Toplanguage (https://groups.google.com/group/pongba/)

I often recommend some books in the toplanguage discussion group, and often ask the cows inside to gather some relevant information, artificial intelligence, machine learning, natural language processing, knowledge discovery (especially, data mining), information retrieval These are undoubtedly the most interesting branches of CS field (also closely related to each other), Here are some of the most recent learning resources related to machine learning and AI in one category:

First of all, there are two great Wikipedia entries, and I'm also a heavy user of Wikipedia, and when I learn something, I often find that it starts with Google in the middle of Wikipedia and ends up in one or several books.

The first is the history of Ai (Artificial Intelligence), which I wrote in the discussion group:

And the article I saw today is the best I have ever seen in Wikipedia. The article is called "The history of Artificial intelligence", along with the AI development timeline, interspersed with countless bull stories, and twists Majestic, "the facts are more surprising than the imagination." Artificial intelligence begins with philosophical speculation, in which there is a stage of help without psychology (especially cognitive neuroscience), only through the induction, introspection, and mathematical tools of the external manifestations of human thought, the most exciting of which is Herbert Simon (father of decision theory, Nobel Prize, Cross-field cattle) wrote an automatic proving machine, which proved the more than 20 theorems of Russell's mathematical principle, one of which is more elegant than the original, and Simon's program uses heuristic search, because the proofs in the axiom system can be simplified to a tree-like search from the condition to the conclusion (but because of the combination explosion, So you have to use heuristic pruning). Later, Simon wrote the GPS (General Problem Solver), which is said to solve some good formalization problems, such as Hanoi. But after all, Simon's research only touches a small, small aspect of human thought--formal Logic, or even narrower deductive reasoning (i.e. not inductive reasoning, transductive Re Asoning (commonly known as analogic thinking). There are many such as Common sense, Vision, especially the most complex Language, consciousness are still unsolved. Another interesting thing is that some people think that the AI problem must be supported by a physical body, a body that can feel the physical rules of the world itself is a powerful source of information, based on this source of information, human beings can keep up with the times to summarize the so-called Common-sense Knowledge (This is the so-called emboddied  mind theory.) ), otherwise like some dude to build common-sense knowledge Base directly, it's silly and naïve to know that people based on the perceptual system from the nature to obtain knowledge is a dynamic automatic updating system, and manually build common sense database is tantamount to the old Expert system practice. Of course, the above only summed up a very small part of my personal feeling is more interesting or novel, everyone sees the interesting place is different, for example, in quite a detailed introduction of neural network theory of the rise and fall. So I strongly suggest you look at yourself again and don't forget the links inside the link to other places.

By the way, Xu 's classmate intends to find time to translate this article, this is a fairly long article, see the E-text waiting to see translation:)

The second one is " ai " (Artificial Intelligence). Of course, there are machine learning and so on. from these items you can find many useful and reliable in-depth references .

Then there are some books

Books:

1. "Programming Collective Intelligence", the introduction of good books in recent years, the cultivation of interest is the most important link, one to see the tome is easy to scare away: P

2. Peter Norvig's "AI, Modern Approach 2nd"(undisputed field Classic).

3. "The Elements of Statistical learning", the math is relatively strong, can do reference.

4. Foundations of statistical Natural Language processing, a recognized classic in the field of natural language processing.

5. "Data Mining, concepts and Techniques", a book written by Chinese scientists, is quite understandable.

6. "Managing gigabytes", a good book of information retrieval.

7. "Information theory:inference and Learning Algorithms", reference books, relatively deep.

Relevant mathematical basis (reference books, not suitable to read through):

1. Linear algebra: This reference book is not listed, many.

2. Matrix Mathematics:"Matrix Analysis", Roger Horn. The undisputed classic of matrix analysis.

3. Probability theory and statistics: "Probability theory and its application", William Feller. is also extremely ox's book, Can the mathematics taste is too heavy, is not suitable does the machine study. So the Du Lei Students in the discussion group recommended all of Statistics and said

In this direction of machine learning, statistics are also very important. Recommend all of statistics, a very concise textbook for CMU, focusing on concepts, simplifying calculations, simplifying concepts and statistical content unrelated to machine learning, which can be said to be a good quick start material.

4. Optimization method:"Nonlinear programming, 2nd" non-linear programming reference. "Convex optimization" convex optimization of the reference book. There are also some books that can be consulted on Wikipedia's most optimized method entries. A deep understanding of the technical details of machine learning methods often requires optimization methods, such as SVM, to pave the way.

Wang Ning students recommended several books:

"machine learning, Tom Michell", 1997.
Old book, Ox Man. Now it seems that the content is not deep, many chapters have the feeling of donuts, but it is suitable for novice (of course, can not "new" to even algorithms and probabilities do not know) get started. For example, the decision Tree section is very exciting, and these years have not been particularly significant progress, so it is not outdated. In addition, this book is a great review of machine learning work for decades 97 years ago, and the list of references is extremely valuable. Domestic have translation and photocopy version, do not know out of print No.

"Modern information retrieval, Ricardo baeza-yates et al". 1999
Old book, Ox Man. It looks like the first book that covers IR completely. Unfortunately, IR has progressed rapidly over the years, and the book is somewhat outdated. It's a good idea to turn over and do a reference. In addition, Ricardo is now the head of Yahoo! for Europe and Latin Ameria.

"Pattern Classification (2ed)", Richard O. Duda, Peter E. Hart, David G. Stork
It's about 01 years old, with a photocopy and color. Not finished, but if you want to learn more about ML and IR, the first three chapters (introduction, Bayesian learning, linear classifier) are compulsory.

There are some classics and I have only one side of the edge, no qualification evaluation. There are also two brochures, the nature of the proceedings, but there are many fronts and details, such as how the index is compressed. Unfortunately forget the name, and I pressed at the bottom of the tank, the next move is even difficult to see the daylight.

(Oh, think of a copy:"Mining the web-discovering knowledge from hypertext Data" )

Say a book of great fame:"Data mining:practical machine learning Tools and techniques". Written by the author of Weka. Unfortunately the content is general. The theoretical part is too thin, and the practical part is very detached from the reality. DM has a lot of introductory books, this one should not be looked at. If you want to learn about Weka, it's good to read the documentation. The second edition has been out, have not read, not clear.

In terms of information retrieval,Du Lei 's classmates re-recommended:

The book on information retrieval is now recommended to see Stanford's "Introduction to Information Retrieval", the book has just been published, the content of course up to date. Another information retrieval the first Daniel Croft is also writing textbooks, should soon be published. It is said to be a very pratical book.

Students interested in information retrieval strongly recommend Dr. Shengcheng's Summer school course at Peking University , which has full slides and reading materials:http://net.pku.edu.cn/~course/cs410/schedule.html

Maximzhao classmate recommended a machine learning:

Add a book: Bishop, "Pattern Recognition and machine learning". There is no photocopy, but it can be down on the Internet. Classics in the classic. Pattern Classification and this book are two must-read books. "Pattern Recognition and machine learning" is very new (07), in Layman's, Shoubushijuan.

Finally, on artificial intelligence (especially, decision-making and judgment), we recommend two interesting books,

One is "simple heuristics, that makes Us Smart"

The other one is "bounded rationality:the Adaptive Toolbox"

Unlike the statistical machine learning approach used in computer science, these two books focus more on the cognitive approach that humans actually use, and the following is a brief description of what I wrote in the discussion group:

The two are written collectively by the German ABC Research Group (an interdisciplinary research group of computer scientists, cognitive scientists, neuroscientists, economists, mathematicians, statisticians, etc.), all of which have aroused widespread interest in the field, especially the previous one, and the latter for Herbert Simon. (The father of decision Science, Nobel Prize winner) proposed the expansion of the human rational model, it can be said that what is the real human intelligence this problem put on the table. The core idea is that our brains simply cannot do a lot of statistical calculations, using Fancy's mathematical methods to explain and predict the world, but rather to confront the uncertain world with simple and robust heuristics (such as the two later very famous heuristics mentioned in the first book: the Re-recognition heuristic (cognition Heuristics) and choose the best. Of course, these two books do not exclude the statistical method is, the data is large when the statistical advantage comes out, and the data is small when the statistical method becomes very bad , the simple heuristic law of mankind to make full use of the regularity in the ecological environment (regularities), are small and robust in terms of computational complexity.

A brief introduction to the second book:

1. Who is Herbert Simon ?

2. What is bounded rationality

3. What this book is about:

I have always felt that human decision-making and judgment is a very fascinating question. The book, in short, can be seen as a more comprehensive and theoretical version of decision and judgment. The systematic and theoretical introduction of various heuristics (heuristics) and their pros and cons in the process of human decision-making and judgment (why they are the quickest and most robust approximation of optimization methods in the case of insufficient information, and why, in some cases, the consequences are bad), For example, Learning machine learning know that naive Bayesian method in many cases is often not worse than Bayesian network, but also fast, such as the higher the dimension of polynomial interpolation is easier to overfit, and the low-order polynomial-based piecewise spline interpolation is proved to be a very robust scheme.

The example mentioned in this book is very interesting: Two teams were sent to design a robot that could catch the thrown baseball on the pitch. The first group made a detailed mathematical analysis and established a fairly complex parabolic approximation model (because of the reasons for the air resistance and so on, so it is not a strict parabola), used to calculate the ball's placement so that the ball can be properly received. Obviously this program is expensive, and the actual operation will take time, we all know that the biological neural network of biological current transmission only hundred meters per second, so computational complexity is a valuable resource for living organisms, so this scheme, although feasible, but not good enough. The second group interviewed the real athletes and listened to how they felt about catching the ball, and then they made a robot: the robot did not do anything at the beginning of the first half of the pitch, until it was close and started running, and kept the eyes from the ball in the running. The latter ensures that the robot's running path must have an intersection with the trajectory of the ball, and that the robot only makes very coarse trajectory estimates throughout the process. Do you know that when you catch a ball, you always stare at it and adjust the direction of movement according to the angle of sight? In fact, that's what humans do, and that's the power of heuristics.

Relative to the psychology and popular science "decision-making and judgment", the book is more theoretical, citing a lot of literature and classic, but also with artificial intelligence and machine learning are cross, there are a lot of mathematical content, the encyclopedia is composed of more than 10 chapters, each chapter is written by different authors, similar to paper , very rigorous, there is no nonsense, and "Psychology of Problem solving" similar. More suitable for geeks reading ha.

In addition, the technical details of the theory can not continue to look at the "decision and judgment" this kind of books (as well as "do not do normal fools" such as the Fool Science Reader), for their own life in the decision-making has a great advantage. Human decision-making and judgment of the use of a lot of heuristics, unfortunately, many of them are in the social environment hundreds of thousands of years ago built up, and not suitable for modern society, so understand these thinking shortcomings, blind spots, to oneself become a good decision-maker has great benefits, And it's a very interesting area in itself.

http://blog.csdn.net/pongba/article/details/2915005

Finish

Machine learning and artificial Intelligence Learning Resource guidance