Machine learning and human

I often use toplanguage
Some books are recommended in the discussion group, and we often ask the ox people to collect relevant information, such as artificial intelligence, machine learning, natural language processing, and Knowledge Discovery (especially Data Mining), Information Retrieval
These are undoubtedly CS
The most interesting branch in the field (also closely related to each other). Here we will classify some learning resources related to machine learning and artificial intelligence recently:

First, there are two very good Wikipedia entries.
Wikipedia
Serious users,When learning a course, we often find that it begins
Wikipedia goes through Google several times
And then stop one or more books.

The first is"History of AI"(History
Of Artificial Intelligence), I wrote in the discussion group:

The article I saw today is in Wikipedia.
I think it's the best. The article titled history of artificial intelligence follows
AI
The development timeline is full of stories, interspersed with countless cool people, and magnificent. It can be said that "facts are even more surprising than imagined ". Artificial Intelligence started with philosophical speculation, and went through a stage without the help of Psychology (especially cognitive neuroscience). It is only through the induction and inner of the external manifestation of human thinking, and the most exciting thing about the exploration of mathematical tools
Herbert Simon
(Father of decision-making theory, Nobel Prize, cross-domain cool-man) an automatic proof machine proves more than 20 theorems in Russell's mathematical principles. One of them is more elegant than the original one, simon
The program uses heuristic search, because proofs in the system can be simplified to tree search from the condition to the conclusion (but due to combination explosion, heuristic pruning is required ). Later
Simon wrote the general problem
Solver), it is said that it can solve some problems that can be well formalized, such as the tower. But after all
Simon
After all, the research only involves a very small aspect of human thinking.
-- Formal logic, or even more narrowly deductive reasoning
(That is, do not include inductive reasoning, transductive Reasoning
(Commonly known as Analogic thinking ). There are many more such as common
Sense, vision, especially the most complex language, Consciousness
There are still unsolved mysteries. Another interesting thing is that some people think
The AI problem must begin with a physical body.
To support this, a body that can feel the world's physical rules is itself a powerful source of information. Based on this source of information, humans can keep up with the times to summarize the so-called
Common-sense knowledge (this is the so-called emboddied mind
Theory. ). Otherwise, it will be built manually like some dude.
Common-sense Knowledge Base
It's silly and naive. Note that people can acquire knowledge from the natural world based on the perception system. It is a dynamic and automatically updated system, and manual knowledge library construction is no different from the old one.
Expert System
. Of course, the above only summarizes a small part of what I personally think is interesting or novel. What everyone sees is different. For example, the rise and fall of neural network theory is described in quite detail. So I strongly recommend that you read it again. Don't forget the link to other places.

By the way,Xu ZheThe student intends to find time to translate this entry. This is a long entry and cannot be viewed.
E text is waiting to see the translation :)

The second is"Artificial Intelligence"(Artificial
Intelligence ). Of course, there areMachine LearningAnd so on.From these entries, We can find many useful and reliable in-depth references..

Then there are some books

Books:

1.Programming collective
Intelligence,In recent years, getting started with a good book is the most important part to cultivate interest. On the top of the page, it is easy to be scared: P

2. Peter norvig'sAI, modern approach
2nd(Classic in a non-controversial domain ).

3.The elements of statistical
Learning,Strong mathematics. You can refer to this document.

4.Foundations of statistical natural language
ProcessingIs recognized as a classic in the natural language processing field.

5.Data mining, concepts and
TechniquesThe books written by Chinese scientists are quite simple.

6.Management
Gigabytes, A good book for information retrieval.

7.Information Theory: inference and learning
AlgorithmsFor more information, see.

Related mathematical BASICS (reference books are not suitable for general reading ):

1. Linear Algebra: This reference book will not be listed.

2. Matrix mathematics:Matrix Analysis, Roger
Horn. Classic in the field of matrix analysis.

3.
Probability Theory and statistics: probability theory and its application, William felle. It is also an excellent book. It can taste too heavy in mathematics and is not suitable for machine learning. So
Du leiRecommended by studentsAll
StatisticsAnd said:

Statistics are equally important in machine learning. Recommended all
Of
Statistics, a concise textbook of CMU, focuses on concepts, simplifies computing, and simplifies Machine
Concepts and statistical content unrelated to learning can be said to be a good Quick Start material.

4. optimization method:Nonlinear Programming,
2ndReference books for nonlinear planning.Convex
OptimizationReference for convex optimization. There are also some books for Reference
Wikipedia
. To thoroughly understand the technical details of machine learning methods, we often need optimization methods (such as SVM.

Wang NingI recommended several books:

Machine learning, Tom Michell,
1997.
Old Book, Niu Ren. Now it seems that the content is not very deep. Many chapters have the feeling of just getting started, but it is very suitable for beginners (of course, it cannot be "new" or even do not know the algorithm and probability. For example, the decision tree is wonderful, and there has been no significant progress in the past few years, so it is not outdated. In addition, this book is a big overview of machine learning over the past 97 years. The reference list is very valuable. Domestic translation and film "restoration and restoration"

Modern information retrieval, Ricardo Baeza-Yates et
Al. 1999
Old Book, Niu Ren. It seems that the first book fully describes IR. It is a pity that IR has made rapid progress over the years, and this book is somewhat outdated. For more information, see. In addition, Ricardo is now a yahoo
Research for Europe and Latin Ameria head.

Pattern Classification (2ed), Richard O.
Duda, Peter E. Hart, David G. Stork
It is also a big block around 01 years old. It has a huge influence on the numbers of "C Huan HUan Xiao Yu L and IR". The first three chapters (Introduction, Bayesian learning, and linear classifier) are required.

There are also some classics that have only one link with me and are not qualified for review. There are also two other books, of the nature of the paper set, which cover a lot of cutting-edge and details, such as how to compress indexes. Unfortunately, I forgot my name and pressed it to the bottom of the box. It was hard to see the sky before the next migration.

(Haha, think of one:Mining the web-
Discovering knowledge from hypertext data)

Let's talk about a famous book:Data Mining: Practical
Machine learning tools and techniques. WEKA
Written by the author. Unfortunately, the content is average. The theoretical part is too thin, and the practical part is very different from the actual part. There are already a lot of entry books for DM, so this book should be available. If you want to learn
WEKA
, Just read the document. The second version has been released and has not been read. It is not clear.

Information retrieval,Du lei
Recommended again:

I suggest reading Stanford's book on Information Retrieval.Introduction
To Information
RetrievalThis book has just been officially published. The content is certainly up.
To
Date. In addition, Master Croft, the first master of information retrieval, is also writing textbooks and should be available soon. It is said to be a very pratical book.

Those who are interested in information retrieval are strongly recommendedDr. Yan chengxiang's summer school courses at Peking UniversityHere are all slides and reading materials:Http://net.pku.edu.cn /~ Course/cs410/schedule.html

Maximzhao
I recommended a machine learning book:

Add a book: Bishop,Pattern Recognition and
Machine Learning.
There is no photocopy, but it can be downloaded online. Classic. Pattern
Classification and this book are two essential books. Pattern
Recognition and Machine
Learning is quite new (). It is easy to understand.

Finally, we recommend two interesting books on AI (especially decision-making and judgment,

One copy isSimple heuristics that makes us
Smart

The other one isBounded Rationality: the adaptive
Toolbox

Different from the statistical machine learning method adopted by computer science, these two books focus more on the cognitive methods actually used by humans. The following is my introduction to the discussion group:

Both are written collectively by the German ABC Research Group (an interdisciplinary research group consisting of computer scientists, cognitive scientists, neuroscientists, economists, mathematicians, and statisticians, it is a book that has aroused widespread attention in the field, especially the previous one.
Herbert Simon
(The father of Decision-making science, the Nobel Prize winner) proposed the expansion of the human rational model.) It can be said that the question of what is real human intelligence has been put on the table. The core idea is that our brain cannot do a lot of statistical computing at all, and it uses fancy's mathematical techniques to explain and predict the world, instead, we face the uncertain world through a simple and robust heuristic method (for example, the first book mentioned two very famous heuristic methods later: recognition of the heuristic method (cognition ).
Heuristics) and
Best ). Of course, these two books do not exclude the statistical methods. When the data volume is large, the statistical advantages come out, and when the data volume is small, the statistical methods becomeVery badThe simple heuristic law of mankind makes full use of the regularity in the ecological environment, so that the computing complexity is small and robust.

Introduction to the second book:

1. Who isHerbert
Simon

2. What isBounded
Rationality

3. What does this book mean:

I have always thought that human decision-making and judgment are a fascinating question. This book can be viewed as a more comprehensive and theoretical version of decision making and judgment. Systematically and theoretically introduce various heuristic methods (heuristics) in human decision making and judgment processes and their advantages and disadvantages
(Why are they fast and robust approximation of optimization methods in the case of insufficient information, and why in some cases it will lead to bad consequences, for example, those who have learned machine learning know that the naive Bayes method is not inferior to the Bayesian Network in many cases, but also fast. For example, the higher the dimension of Polynomial Interpolation, the more likely it is to overfit, however, piecewise spline interpolation based on low degree polynomials proves to be a very robust solution ).

The example mentioned in this book is very interesting: two teams are sent to design a robot that can catch the baseball players on the court. The first group made a detailed mathematical analysis and established a rather complex parabolic approximation model (because the air resistance and other factors are also considered, so it is not a strict parabolic curve), which is used to calculate the ball placement, so that the ball can be correctly received. Obviously, this solution is costly, and the actual computation also takes time. We all know that the biological current transmission in the biological neural network is only within one hundred meters per second, so
Computational Complexity
Biology is a valuable resource, so this solution is feasible but not good enough. The second group interviewed real athletes and listened to their feelings about how they caught the ball. Then they made such a robot: this robot does nothing at the First Half of the ball's throw. It does not start to run until it is close, and keeps the angle of view between the eyes and the ball unchanged during the running, the latter ensures that the running route of the robot is always at the intersection of the ball's trajectory. during the whole process, the robot only makes rough track estimation. Do you always stare at the ball when you catch the ball, and then adjust the running direction based on the line of sight? In fact, this is what humans do.
The power of heuristics.

This book is more theoretical than the decision-making and judgment, which is biased towards psychology and popular science. There are also a lot of references and classics, and it is also different from artificial intelligence and machine learning, there is also a lot of mathematical content in it. The book consists of more than a dozen chapters, each of which is written by different authors, similar
Paper is the same, rigorous, and nonsense.
Of problem solving. It is suitable for reading geeks.

In addition, if you cannot read the technical details of the theory, you are advised to read the books such as decision making and judgment (and other silly books such as don't be a normal dummies ), it is of great benefit to making decisions in your life. Humans use a lot
Heuristics
Unfortunately, many of them have been established in the social environment of 100,000 years ago and are not suitable for modern society. Therefore, they understand the shortcomings and blind spots in these thinking, it is of great benefit to become a good decision maker, and it is also a very interesting field.