In the summer of 1956, an academic conference held at Dartmouth University in the United States was recognized as the starting point for global artificial intelligence research for many years. In the spring of 2016, an AlphaGo battled with Li Shishi, the world's top Go master, to push the world into a new wave of artificial intelligence.
After two ups and downs, artificial intelligence is now entering the eve of the global outbreak. In China alone, hundreds of millions of people watched AlphaGo and Li Shishi directly or indirectly. In early 2016, IBM also promoted globally based on IBM Watson's cognitive computing. Watson's predecessor defeated chess in 1997. Master Kasparov's "dark blue".
There is an old saying in China called 60 years of reincarnation. However, for artificial intelligence, the next 60 years are not only reincarnation, but new life. The artificial intelligence process of the first 60 years can be described as “infinite movementâ€; the development of artificial intelligence in the last 60 years can be expected with “infinityâ€.
"Infinity Movement" is an Italian violin song. Xing Bo, a professor of artificial intelligence at Carnegie Mellon University, named the new generation of distributed machine learning system developed by his research team. "Infinity" is another kind of meaning, representing the never-ending imagination of intelligent machines and the impulse to practice in ancient times in the past 60 years and even further.
People’s imagination of intelligent machines is endless
Aristotle once said that if the machine can do a lot of work, you can't let humans liberate. Science fiction movies such as "Star Wars", "The Matrix" and "Artificial Intelligence" have inspired generations of scholars and industrialists to devote themselves to the study of artificial intelligence. The main inventor of the AlphaGo algorithm was influenced by the "dark blue" and joined the ranks of AI.
In the development of the first 60 years, artificial intelligence research has also achieved phased results. In particular, supervised deep learning has developed into a mature stage in the field of artificial intelligence such as natural language understanding, speech recognition, and image recognition. Next, it is the future of unsupervised deep learning pioneered by AlphaGo – from the enhanced learning of human “supervisionâ€.
In fact, with the invention of computers, people have been discussing what kind of artificial intelligence will this lead to? One type of foresight is the ability to produce functional artificial intelligence, which is the result of today's supervised deep learning. Another point of view is that artificial intelligence can imitate people's thinking and emotional activities, which is the future that unsupervised deep learning will create.
When intelligent machines can open their eyes to see the world and acquire intelligence through independent exploration of the world, the possible changes in the future are "infinity." From "infinite movement" to "infinity", 2016 is destined to be a wonderful year.
Part I: The rhythm of "infinite movement" in the first 60 years
The 60-year development of artificial intelligence is the infinite rhythm between ups and downs, cold winters and new trends, disappointment and hope, looking for the best combination of theory and practice.
After graduating from Tsinghua University, Xing Bo went to Rutgers University and Berkeley to pursue postgraduate studies, and then became a professor in the field of artificial intelligence at Carnegie Mellon University. Carnegie Mellon is an important research base for artificial intelligence in the world, and many original achievements are from this university.
Xing Bo, Professor of Artificial Intelligence, Carnegie Mellon University
Xing Bo established an artificial intelligence group SAILING LAB at Carnegie Mellon University, trying to make breakthroughs in all aspects of artificial intelligence. Theoretical research includes maximum likelihood and maximum interval learning of probability map models, nonparametric space high dimensional reasoning, and instability. State time series analysis, nonparametric Bayesian inference, etc., applied research including computational biology, population genetics, genomics, social networks and social groups, Internet-level text mining and natural language processing, computational finance, etc.
On March 19th, 2016, on the 4th day after AlphaGo defeated Li Shishi, Xing Bo entered the Jingdong Group with the understanding of the Future Forum. Xing Bo reviewed the global process of artificial intelligence. As a science and engineering field, artificial intelligence has benefited from the intersection of many scientific developments such as international science, computer science, information theory and cybernetics in the 20th century. The study of artificial intelligence is based on a very basic assumption that human thinking activities can be replaced mechanically.
Global Artificial Intelligence Conference 60 years ago
When it comes to artificial intelligence, you can't help but mention the originator: Turing. In 1936, the British mathematician and logician Alan Messon Turing (1912~1954) proposed an abstract computational model, TuringMachine, which used paper-based machines to simulate people's mathematical operations. The process, Turing himself is regarded as the father of computer science.
In 1959, Turing published an epoch-making paper "Computers and Intelligence", which proposed the famous Turing test in the field of artificial intelligence - if the computer can answer a series of questions raised by human testers within 5 minutes, And more than 30% of the answers make the tester mistakenly believe that it is answered by humans, then the computer passes the test and can conclude that the machine is intelligent.
Carnegie Mellon University
The concept of Turing test greatly influences the definition of function of artificial intelligence. In this way, Carnegie Mellon's "Logical Theorist" program by two scientists A. Newell and H. Simon proves Russell's "Mathematical Principles" very subtly 38 of the 52 lanes. Simon claims that within 10 years, the machine can reach the same height as human intelligence.
The first batch of artificial intelligence explorers found a common language. In 1956, 60 years ago, they held a meeting at Dartmouth University in the United States, hoping to establish artificial intelligence as a scientific task and a complete path. Participants also claimed that the characteristics of artificial intelligence can be accurately described, and can be simulated and implemented by machine after accurate description. It was later widely believed that the Dartmouth meeting marked the official birth of artificial intelligence.
Artificial intelligence first wave and winter
The Dartmouth Conference promoted the world's first wave of artificial intelligence, from 1956 to 1974. At that time, the optimistic atmosphere filled the entire academic world. There were many world-class inventions in the algorithm, including a prototype called enhanced learning (Berman formula). Enhanced learning is the core idea of ​​Google AlphaGo algorithm. The deep learning model that is often heard now, the prototype is called the perceptron, and was invented in those years.
Dartmouth University 60 years ago
In addition to new advances in algorithms and methodologies, in the first wave, scientists also created clever machines. Among them, a machine called STUDENT (1964) can prove the application problem, and a machine called ELIZA (1966) can realize simple man-machine dialogue. Therefore, the artificial intelligence community believes that artificial intelligence can really replace human beings at such a speed of development.
The first artificial intelligence winter appeared between 1974 and 1980. What is going on here? Because people find that logic provers, perceptrons, reinforcement learning, etc. can only do very simple, very specialized and very narrow tasks, and can't cope with a little out of scope. There are two limitations in this respect: on the one hand, the mathematical models and mathematical methods on which artificial intelligence is based are found to have certain defects; on the other hand, there are many computational complexity that increase exponentially, so it becomes an impossible calculation task. .
Congenital defects lead to artificial bottlenecks in the early development process, so the first winter is coming soon, and the funding for artificial intelligence is correspondingly reduced or cancelled.
Modern PC "promotes" the second artificial intelligence winter
In the 1980s, Carnegie Mellon University created an expert system (1980) for DEC, which helped DEC save about $40 million a year, especially in decision making. . Encouraged by this, many countries, including Japan and the United States, have once again invested heavily in the development of the so-called 5th generation computer (1982), which was called artificial intelligence computer.
In the 1980s, major inventions of artificial intelligence mathematical models emerged, including the famous multi-layer neural network (1986) and the BP back-propagation algorithm (1986). There were also highly intelligent machines that could play chess with humans (1989). ). In addition, other achievements include machines that automatically recognize the zip code on the envelope, which is achieved through an artificial intelligence network with an accuracy of over 99%, which has exceeded the level of ordinary people. So, everyone began to think that artificial intelligence still has a play.
Early expert system Symbolics 3640
However, the emergence of modern PCs from 1987 to 1993 allowed the winter of artificial intelligence to come again. At that time, Apple and IBM began to promote the first generation of desktop computers, and the computer began to enter the personal home, which cost far less than the machines such as Symbolics and Lisp used by the expert system. Compared to modern PCs, expert systems are considered old and very difficult to maintain. As a result, government funding began to fall, and winter was coming again.
At that time, even scholars were not too embarrassed to say that they were engaged in artificial intelligence research. People began to think about where artificial intelligence is going, and what kind of artificial intelligence to achieve.
The dawn of modern AI: new tools, new ideas and Moore's Law
How to do useful things with limited resources is a constant challenge of artificial intelligence. A realistic approach is to use engineering methods to simplify functions, deploy simple mathematical models, and develop powerful aircraft engines, just like humans make airplanes, inspired by the biological world.
The dawn of modern AI took place at this stage, with new mathematical tools, new theories, and Moore's Law. Artificial intelligence is also determining its own direction. One of the choices is to make practical and functional artificial intelligence, which leads to a new artificial intelligence path. Due to the clarity and simplification of artificial intelligence tasks, new prosperity has been brought about.
Schematic diagram of deep learning algorithm based on neural network
In terms of new mathematical tools, mathematical models that existed in the literature of mathematics or other disciplines were re-explored or invented. A number of significant achievements at the time, including the recent Turing Award-winning graph model and graph optimization, deep learning networks, etc., were re-raised about 15 years ago and re-started research.
In the new theory, due to the simplification of the natural world by mathematical models, there is very clear mathematical logic, which makes theoretical analysis and proof possible, and can analyze how much data and calculations are needed to obtain the desired results. It is very helpful to develop a corresponding computing system.
On a more important aspect, Moore's Law makes computing more powerful, and powerful computers are rarely used in early research on artificial intelligence, because early artificial intelligence research was more often defined as mathematical and algorithmic research. When more powerful computing power is transferred to artificial intelligence research, the research effect of artificial intelligence is significantly improved.
Due to this series of breakthroughs, artificial intelligence has created a new boom. The earliest result was the 1997 IBM Deep Blue victory over chess masters. In terms of more general-purpose functionality, the machine can meet or exceed human standards in mathematical competitions and in the recognition of pictures.
The boom in artificial intelligence has also contributed to the advancement of robotics, including the use of artificial intelligence principles in the design of robotic dogs. Whether it is an artificial intelligence dog or an unmanned vehicle driving, it is not written in a programming method. Instead, it uses a set of learning algorithms to continuously walk and drive in the simulator, allowing the machine to generate its own behavioral strategy. This is artificial intelligence and original control. On the most different places.
2011, Facebook Challenge
In 2011, Xing Bo ushered in the first academic leave of a professor, and American professors could take a vacation every six years. Xing Bo chose to go to a very young company as a visiting professor. This was Facebook at the time. At that time, only 500 people of Facebook set up their own lab in the warehouse of Stanford University. At that time, Facebook proposed to connect hundreds of millions of users, and also hoped to use artificial intelligence to deliver valuable advertisements to increase the company's revenue.
Facebook's goal at the time was to increase the number of users from 100 million to 1 billion in the near future. Xing Bo's mission is to help Facebook achieve this vision. As Facebook's first visiting professor, his first task was to connect users to social networks and then project that connection into social space for community detection and community detection to implement users. Grouping and characterization.
This task is not difficult, it can be achieved by mixing the member random block model, which is the best AI algorithm for processing network data in 2011. But there is a problem, that is, the computational complexity is square-level, that is, 100 times CPU and storage are required for every 10 times increase in the number of users, so the maximum processing of 10,000 people per single machine is the biggest problem at that time.
Xing Bo then achieved computational acceleration by studying the algorithm model, including the feature that is more powerful in the social network extraction than the "edge" is called "triangle", and the model is also upgraded from the hybrid block model to the hybrid triangle model. Hybrid algorithms have achieved significant innovations, and computational complexity is declining. The research results at that time were used in the global movie star network research. The network of about 1 million people can show people in real time that they are constantly driving in social space to find friends and fall into different social groups.
KNB1-100 Miniature Circuit Breaker
KNB1-100 Mini Circuit breakers, also named as the air switch which have a short for arc extinguishing device. It is a switch role, and also is a automatic protection of low-voltage electrical distribution. Its role is equivalent to the combination of switch. Fuse. Thermal Relay and other electrical components. It mainly used for short circuit and overload protection. Generally, According to the poles, mini Circuit breaker can be divided into 1P , 1P+N , 2P, 3P and 4P.
KNB1-100 Miniature Circuit Breaker,Electronics Miniature Circuits Breaker,Automatic Miniature Circuit Breaker,Mini Circuit Breaker
Wenzhou Korlen Electric Appliances Co., Ltd. , https://www.zjmoldedcasecircuitbreaker.com