The 18th century Bayesian theorem became the new force of Google calculation
Author: Michael Kanellos
Thomas Bayes, a great mathematics master, his theory illuminates today's calculation field, different from his colleagues: he believes that God's existence can prove through the equation, his most important work is released by others, and he already Died in 241 years.
The 18th century priests have become part of the mathematical foundation of the application development.
Searching for giants Google and Autonomy, a company for sale Information Recovery Tools, using Bayesian Principles, provides approximation (but technology is not exact) results. Researchers also use the Bayesian model to judge the relationship between symptoms and diseases, create personal robots, and develop artificial intelligence devices that can determine action based on data and experience.
Although it sounds well, this principle means - roughly - it is very simple: some probability of things can be approximately estimated by the frequency that it has occurred in the past. The researchers apply this principle on everything, from gene research to filter email.
A detailed mathematical summary can be found on the University of Minnesota. A Bayes Rule Applet program on Gametheory.net allows you to answer questions such as "if you test some disease, how big risks".
A famous advocate in Bayesian theory is Microsoft. The company uses probability to use its Notification Platform. This technology will be built into Microsoft's future software, and allows computer and cellular phones to automatically filter information, do not require user assistance, automatic planning conference and contact with others.
If successful, the technology will lead to "Context Server" - an electronic housekeeper's appearance, it can explain people's daily habits and organize their lives in an incurred environment.
"Bayes' research is used to determine how I should best allocate calculations and bandwidth," Eric Horv
IT
Z said he is a senior researcher and group manager of the Microsoft Research Department Adaptive Systems & Interaction Group. "I personally believe in this uncertain world, you can't know everything, and probability is the foundation of any intelligence."
By the end of this year, Intel will also release its own Bayesian-based toolkit. A experimental warning doctor that the doctor said that the patient may suffer from suffering. This development will be discussed on this week later in the company's Developer Forum.
Although it is very popular today, Bayes' theory is not always widely accepted: 10 years ago, Bayes researchers were still in front of them. However, afterwards, the improved mathematical model, the effective results of the faster computer and experiment increase the new credibility of this school.
"One of the problems is that it is over-propagated," Intel Microprocessor Laboratory Application Software and Technology Management Manager Omid Moghadam. "In fact, the ability to deal with anything does not exist. Real implementation happened in the past 10 years."
Bayes Duted Bayes Theory can be roughly briefly described as a principle: in order to foresee the future, you must look at it. Bayes' Theory indicates that the probability of anything in the future can be estimated by calculating the frequency of it in the past. What is the probability of a positive face positive? Experimental data indicates that this value is 50%.
"Bayes said in essentially, everything has uncertainty, you have different probability types," said the professor of Stanford's Management Science and Engineering Department (Department of Management, Engineering, Ron Howard).
For example, suppose is not a coin, a researcher throws the plastic push up, and wants to see how big the probability of the staple floor is, or how many likelihood is the side, and the nail is pointing in. In shape, errors, weight distributions, and other factors in the formation process will affect the result. The appeal of Bayes technology is its simplicity. The prediction is completely dependent on the collected data - the more data obtained, the better the result. Another advantage is that the Bayes model can be self-corrected, that is, the data changes, and the results will follow.
The ideology of probability changes the way people and computer interactions. "This idea is that the computer can more like a helper, not just a terminal device," Peter Norvig is said. He is Google's security quality. He said, "You are looking for some guidance, not a standard answer."
From this transformation, the study benefits are not shallow. A few years ago, the so-called Boolean search engine usually requires the search to submit the search according to the syntax of "if, and or but", then look for the word. The current search engine uses a complex algorithm to search for the database and find out possible matching.
As shown in the example of the pushpiece, the complexity and the need for more data may grow. Due to the emergence of powerful computers, it is possible to control the results necessary to transition to the approximation of the approximation.
More importantly, researchers such as UCLA's Judea Pearl studied how to make Bayes models can better track the conditional relationship between different phenomena, which can greatly reduce the amount of calculation.
For example, a large-scale survey of lung cancer for a large scale may find that it is a non-extensive disease, but some relevance may be found if it is limited to investigations within the smokers. Checking patients with lung cancer can help investigate the relationship between custom habits and this disease.
"Each individual property or sign may depend on many different things, but directly determine it is a few things, the Assistant Professor of the Standfrove Computer Science Department (Computer Science Department, said Daphne Koller. "In the past 15 years, people have reform in terms of tools, which allows you to describe a lot of people."
As with other items, Koller is a probability theory technology to better link the disease and the disease and link genetic genes and specific cell phenomena.
Record a related technique called Hidden Markov model, allowing probability to predict the order. For example, a speech recognition application knows that the letter often after "Q" is "U". In addition to these, the software can also calculate the probability of appearing in "QAGGA" (a name of extinct zebra).
Probability technology has been built in Microsoft's products. Outlook Mobile Manage is a software that can decide when to issue a parallel parallel panel. It is from prior
IT
IES is developing, PRIOR
IT
IES is an experimental system that Microsoft announced in 1998. The troubleshooting engine of Windows XP is also dependent on probability.
As the company's NOTIFICATION Platform starts built into the product, there will be more application software in the next year, Microsoft's HORV
IT
z is indicated like this.
An important component of Notification Platform is called CORDINATE, which collects data from a personal calendar, keyboard, sensor camera, and other sources to understand a person's life and habits. The data collected may include time, working hours, and lunch time length, which type of phone or email is saved, and which information is deleted, the frequency of the keyboard is used in a certain day, and so on.
These data can be used to manage other information received by the information flow and the user. For example, if a manager sends an e-mail to an employee at 2:40 in the afternoon, Coordinate can check the employee's calendar, and find that he has a meeting at 2:00 pm. The program can also scan data about the employee habit and then find that the employee usually reuse the keyboard approximately one hour after a meeting. The program may also find that the employee usually responds to the manager's email within 5 minutes. According to these data, the software can estimate that the employee may not respond to the email at least 20 minutes, which may send this information to the employee's mobile phone. At the same time, the software may decide not to forward someone else's email. "We are balanced to distinguish your value for the value of information," HORV
IT
z representation. Using this software, he insisted on, "can let more people keep up with things to develop, not being overwhelmed by a lot of information."
Horv
IT
Z Supplement, Privacy, and user control for these functions are determined. The caller does not know why a message may be prioritized or postponed.
Microsoft also uses the Bayes model on some of the other products, including DeepListener, Quartet, Smartoof, and TimeWave. Consumer multimedia software is also not shallow, HORV
IT
z representation.
Bayes technology is not only applied in the PC field. In Univers
IT
Y 'Rochester, researchers found a person's pace can change before one step. Although this change is too subtle for humans, a camera connected together can capture and track this action. If you walk abnormally, your computer can issue an alert.
A experimental safety camera adopts the same principle: most people who arrive at the airport will directly go directly to the destination, so if someone stops the car, then it is not normal to go to another, so it may cause alert. . This fall, a basic engine that creates a Bayes model and technical information will be announced on the developer website of Intel.
Although the technology is simple and easy to understand, the calculation of it may be slower. Horv
IT
Z recalls that he is one of the only two probabilities and artificial intelligence graduates in the 1980s. All other people learn the logic system, which is the mode and world interaction of "if and then".
"The probability is not popular at that time," HORV
IT
z representation. However, when the logic system is not able to predict all accidents, the trend has changed.
Many researchers have begun to recognize that human decision-making process is much mysterious than originally imagined. "There is a cultural prejudice in the field of artificial intelligence," Koller is expressed. "People now admit that they don't know how their brains work."
Even in his era, Bayes found himself outside the mainstream. He was born in London in 1702, and he became a presbyterian minister. Although he saw his two papers were published, his theory was very effective, but "Essay Toward Solving A Problem in the DoCtrine of Chances" has been to his third year after death, that is, 1764 was published .
His court members have always been a mystery, until recently, some letters of new discovery indicate that his private and other ideas in England consistent.
"As I know, he never wrote the Bayesian theorem," HoWard said.
Pierre Simon Laplace in Richard Price and France became early supporters. The theory and later George Boole, the father of the Boolean mathematics, and the theory of the theory. George Boole's theory is based on algebraic logic and eventually leads to the birth of the binary system. It is also one of the Boole, one of the royal members, died in 1864. Although the importance of probability is not suspicious, but the debate about its application has not stopped. Critics periodically claim that the Bayes model relies on subjective data, and human beings will judge whether the answer is correct. The probability discussion model does not completely solve the problem of fine difference in human thinking.
"How does children learn is still not very clear now," said the IBM Research Department's Scientific and Software Vice President Alfred Spector. He plans to combine statistical methods and logic systems in his Combination Hypothesis. "I originally believed that it was a scope of statistics, but from a certain aspect, you will find not only statistical problems."
However, it is very likely that probability is the foundation.
"This is a foundation," HORV
IT
z representation. "It is ignored for a while, but it is the foundation of reasoning."
